24 #include "imageclasses/ImgReaderGdal.h"
25 #include "imageclasses/ImgWriterGdal.h"
26 #include "imageclasses/ImgReaderOgr.h"
27 #include "imageclasses/ImgWriterOgr.h"
28 #include "base/Optionpk.h"
29 #include "base/PosValue.h"
30 #include "algorithms/ConfusionMatrix.h"
31 #include "algorithms/svm.h"
119 enum SVM_TYPE {C_SVC=0, nu_SVC=1,one_class=2, epsilon_SVR=3, nu_SVR=4};
120 enum KERNEL_TYPE {linear=0,polynomial=1,radial=2,sigmoid=3};
123 #define Malloc(type,n) (type *)malloc((n)*sizeof(type))
127 int main(
int argc,
char *argv[])
129 vector<double> priors;
133 Optionpk<string> training_opt(
"t",
"training",
"Training vector file. A single vector file contains all training features (must be set as: b0, b1, b2,...) for all classes (class numbers identified by label option). Use multiple training files for bootstrap aggregation (alternative to the bag and bsize options, where a random subset is taken from a single training file)");
135 Optionpk<string> label_opt(
"label",
"label",
"Attribute name for class label in training vector file.",
"label");
136 Optionpk<unsigned int> balance_opt(
"bal",
"balance",
"Balance the input data to this number of samples for each class", 0);
137 Optionpk<bool> random_opt(
"random",
"random",
"Randomize training data for balancing and bagging",
true, 2);
138 Optionpk<int> minSize_opt(
"min",
"min",
"If number of training pixels is less then min, do not take this class into account (0: consider all classes)", 0);
139 Optionpk<unsigned short> band_opt(
"b",
"band",
"Band index (starting from 0, either use band option or use start to end)");
142 Optionpk<double> offset_opt(
"offset",
"offset",
"Offset value for each spectral band input features: refl[band]=(DN[band]-offset[band])/scale[band]", 0.0);
143 Optionpk<double> scale_opt(
"scale",
"scale",
"Scale value for each spectral band input features: refl=(DN[band]-offset[band])/scale[band] (use 0 if scale min and max in each band to -1.0 and 1.0)", 0.0);
144 Optionpk<double> priors_opt(
"prior",
"prior",
"Prior probabilities for each class (e.g., -p 0.3 -p 0.3 -p 0.2 ). Used for input only (ignored for cross validation)", 0.0);
145 Optionpk<string> priorimg_opt(
"pim",
"priorimg",
"Prior probability image (multi-band img with band for each class",
"",2);
147 Optionpk<string> cmformat_opt(
"cmf",
"cmf",
"Format for confusion matrix (ascii or latex)",
"ascii");
148 Optionpk<std::string> svm_type_opt(
"svmt",
"svmtype",
"Type of SVM (C_SVC, nu_SVC,one_class, epsilon_SVR, nu_SVR)",
"C_SVC");
149 Optionpk<std::string> kernel_type_opt(
"kt",
"kerneltype",
"Type of kernel function (linear,polynomial,radial,sigmoid) ",
"radial");
151 Optionpk<float> gamma_opt(
"g",
"gamma",
"Gamma in kernel function",1.0);
152 Optionpk<float> coef0_opt(
"c0",
"coef0",
"Coef0 in kernel function",0);
153 Optionpk<float> ccost_opt(
"cc",
"ccost",
"The parameter C of C_SVC, epsilon_SVR, and nu_SVR",1000);
154 Optionpk<float> nu_opt(
"nu",
"nu",
"The parameter nu of nu_SVC, one_class SVM, and nu_SVR",0.5);
155 Optionpk<float> epsilon_loss_opt(
"eloss",
"eloss",
"The epsilon in loss function of epsilon_SVR",0.1);
156 Optionpk<int> cache_opt(
"cache",
"cache",
"Cache memory size in MB",100);
157 Optionpk<float> epsilon_tol_opt(
"etol",
"etol",
"The tolerance of termination criterion",0.001);
158 Optionpk<bool> shrinking_opt(
"shrink",
"shrink",
"Whether to use the shrinking heuristics",
false);
159 Optionpk<bool> prob_est_opt(
"pe",
"probest",
"Whether to train a SVC or SVR model for probability estimates",
true,2);
161 Optionpk<unsigned short> comb_opt(
"comb",
"comb",
"How to combine bootstrap aggregation classifiers (0: sum rule, 1: product rule, 2: max rule). Also used to aggregate classes with rc option.",0);
163 Optionpk<int> bagSize_opt(
"bagsize",
"bagsize",
"Percentage of features used from available training features for each bootstrap aggregation (one size for all classes, or a different size for each class respectively", 100);
164 Optionpk<string> classBag_opt(
"cb",
"classbag",
"Output for each individual bootstrap aggregation");
165 Optionpk<string> mask_opt(
"m",
"mask",
"Only classify within specified mask (vector or raster). For raster mask, set nodata values with the option msknodata.");
166 Optionpk<short> msknodata_opt(
"msknodata",
"msknodata",
"Mask value(s) not to consider for classification (use negative values if only these values should be taken into account). Values will be taken over in classification image.", 0);
169 Optionpk<string> oformat_opt(
"of",
"oformat",
"Output image format (see also gdal_translate).",
"GTiff");
170 Optionpk<string> option_opt(
"co",
"co",
"Creation option for output file. Multiple options can be specified.");
171 Optionpk<string> colorTable_opt(
"ct",
"ct",
"Color table in ASCII format having 5 columns: id R G B ALFA (0: transparent, 255: solid)");
173 Optionpk<string> entropy_opt(
"entropy",
"entropy",
"Entropy image (measure for uncertainty of classifier output",
"",2);
174 Optionpk<string> active_opt(
"active",
"active",
"Ogr output for active training sample.",
"",2);
175 Optionpk<string> ogrformat_opt(
"f",
"f",
"Output ogr format for active training sample",
"SQLite");
178 Optionpk<short> classvalue_opt(
"r",
"reclass",
"List of class values (use same order as in class opt).");
181 oformat_opt.setHide(1);
182 option_opt.setHide(1);
184 bstart_opt.setHide(1);
186 balance_opt.setHide(1);
187 minSize_opt.setHide(1);
189 bagSize_opt.setHide(1);
191 classBag_opt.setHide(1);
193 priorimg_opt.setHide(1);
194 offset_opt.setHide(1);
195 scale_opt.setHide(1);
196 svm_type_opt.setHide(1);
197 kernel_type_opt.setHide(1);
198 kernel_degree_opt.setHide(1);
199 coef0_opt.setHide(1);
201 epsilon_loss_opt.setHide(1);
202 cache_opt.setHide(1);
203 epsilon_tol_opt.setHide(1);
204 shrinking_opt.setHide(1);
205 prob_est_opt.setHide(1);
206 entropy_opt.setHide(1);
207 active_opt.setHide(1);
208 nactive_opt.setHide(1);
209 random_opt.setHide(1);
211 verbose_opt.setHide(2);
215 doProcess=training_opt.retrieveOption(argc,argv);
216 input_opt.retrieveOption(argc,argv);
217 output_opt.retrieveOption(argc,argv);
218 cv_opt.retrieveOption(argc,argv);
219 cmformat_opt.retrieveOption(argc,argv);
220 tlayer_opt.retrieveOption(argc,argv);
221 classname_opt.retrieveOption(argc,argv);
222 classvalue_opt.retrieveOption(argc,argv);
223 oformat_opt.retrieveOption(argc,argv);
224 ogrformat_opt.retrieveOption(argc,argv);
225 option_opt.retrieveOption(argc,argv);
226 colorTable_opt.retrieveOption(argc,argv);
227 label_opt.retrieveOption(argc,argv);
228 priors_opt.retrieveOption(argc,argv);
229 gamma_opt.retrieveOption(argc,argv);
230 ccost_opt.retrieveOption(argc,argv);
231 mask_opt.retrieveOption(argc,argv);
232 msknodata_opt.retrieveOption(argc,argv);
233 nodata_opt.retrieveOption(argc,argv);
235 band_opt.retrieveOption(argc,argv);
236 bstart_opt.retrieveOption(argc,argv);
237 bend_opt.retrieveOption(argc,argv);
238 balance_opt.retrieveOption(argc,argv);
239 minSize_opt.retrieveOption(argc,argv);
240 bag_opt.retrieveOption(argc,argv);
241 bagSize_opt.retrieveOption(argc,argv);
242 comb_opt.retrieveOption(argc,argv);
243 classBag_opt.retrieveOption(argc,argv);
244 prob_opt.retrieveOption(argc,argv);
245 priorimg_opt.retrieveOption(argc,argv);
246 offset_opt.retrieveOption(argc,argv);
247 scale_opt.retrieveOption(argc,argv);
248 svm_type_opt.retrieveOption(argc,argv);
249 kernel_type_opt.retrieveOption(argc,argv);
250 kernel_degree_opt.retrieveOption(argc,argv);
251 coef0_opt.retrieveOption(argc,argv);
252 nu_opt.retrieveOption(argc,argv);
253 epsilon_loss_opt.retrieveOption(argc,argv);
254 cache_opt.retrieveOption(argc,argv);
255 epsilon_tol_opt.retrieveOption(argc,argv);
256 shrinking_opt.retrieveOption(argc,argv);
257 prob_est_opt.retrieveOption(argc,argv);
258 entropy_opt.retrieveOption(argc,argv);
259 active_opt.retrieveOption(argc,argv);
260 nactive_opt.retrieveOption(argc,argv);
261 verbose_opt.retrieveOption(argc,argv);
262 random_opt.retrieveOption(argc,argv);
264 catch(
string predefinedString){
265 std::cout << predefinedString << std::endl;
270 cout <<
"Usage: pksvm -t training [-i input -o output] [-cv value]" << endl;
272 std::cout <<
"short option -h shows basic options only, use long option --help to show all options" << std::endl;
276 if(entropy_opt[0]==
"")
278 if(active_opt[0]==
"")
280 if(priorimg_opt[0]==
"")
281 priorimg_opt.clear();
284 std::map<std::string, svm::SVM_TYPE> svmMap;
286 svmMap[
"C_SVC"]=svm::C_SVC;
287 svmMap[
"nu_SVC"]=svm::nu_SVC;
288 svmMap[
"one_class"]=svm::one_class;
289 svmMap[
"epsilon_SVR"]=svm::epsilon_SVR;
290 svmMap[
"nu_SVR"]=svm::nu_SVR;
292 std::map<std::string, svm::KERNEL_TYPE> kernelMap;
294 kernelMap[
"linear"]=svm::linear;
295 kernelMap[
"polynomial"]=svm::polynomial;
296 kernelMap[
"radial"]=svm::radial;
297 kernelMap[
"sigmoid;"]=svm::sigmoid;
299 assert(training_opt.size());
301 if(verbose_opt[0]>=1){
303 std::cout <<
"input filename: " << input_opt[0] << std::endl;
305 std::cout <<
"mask filename: " << mask_opt[0] << std::endl;
306 std::cout <<
"training vector file: " << std::endl;
307 for(
int ifile=0;ifile<training_opt.size();++ifile)
308 std::cout << training_opt[ifile] << std::endl;
309 std::cout <<
"verbose: " << verbose_opt[0] << std::endl;
311 unsigned short nbag=(training_opt.size()>1)?training_opt.size():bag_opt[0];
312 if(verbose_opt[0]>=1)
313 std::cout <<
"number of bootstrap aggregations: " << nbag << std::endl;
323 bool maskIsVector=
false;
326 extentReader.open(mask_opt[0]);
328 readLayer = extentReader.getDataSource()->GetLayer(0);
329 if(!(extentReader.getExtent(ulx,uly,lrx,lry))){
330 cerr <<
"Error: could not get extent from " << mask_opt[0] << endl;
334 catch(
string errorString){
340 if(active_opt.size()){
341 prob_est_opt[0]=
true;
343 activeWriter.open(active_opt[0],ogrformat_opt[0]);
344 activeWriter.createLayer(active_opt[0],trainingReader.getProjection(),wkbPoint,NULL);
345 activeWriter.copyFields(trainingReader);
347 vector<PosValue> activePoints(nactive_opt[0]);
348 for(
int iactive=0;iactive<activePoints.size();++iactive){
349 activePoints[iactive].value=1.0;
350 activePoints[iactive].posx=0.0;
351 activePoints[iactive].posy=0.0;
354 unsigned int totalSamples=0;
355 unsigned int nactive=0;
356 vector<struct svm_model*> svm(nbag);
357 vector<struct svm_parameter> param(nbag);
364 if(priors_opt.size()>1){
365 priors.resize(priors_opt.size());
367 for(
short iclass=0;iclass<priors_opt.size();++iclass){
368 priors[iclass]=priors_opt[iclass];
369 normPrior+=priors[iclass];
372 for(
short iclass=0;iclass<priors_opt.size();++iclass)
373 priors[iclass]/=normPrior;
378 if(bstart_opt.size()){
379 if(bend_opt.size()!=bstart_opt.size()){
380 string errorstring=
"Error: options for start and end band indexes must be provided as pairs, missing end band";
384 for(
int ipair=0;ipair<bstart_opt.size();++ipair){
385 if(bend_opt[ipair]<=bstart_opt[ipair]){
386 string errorstring=
"Error: index for end band must be smaller then start band";
389 for(
int iband=bstart_opt[ipair];iband<=bend_opt[ipair];++iband)
390 band_opt.push_back(iband);
395 cerr << error << std::endl;
400 std::sort(band_opt.begin(),band_opt.end());
402 map<string,short> classValueMap;
403 vector<std::string> nameVector;
404 if(classname_opt.size()){
405 assert(classname_opt.size()==classvalue_opt.size());
406 for(
int iclass=0;iclass<classname_opt.size();++iclass)
407 classValueMap[classname_opt[iclass]]=classvalue_opt[iclass];
412 vector< vector<double> > offset(nbag);
413 vector< vector<double> > scale(nbag);
414 map<string,Vector2d<float> > trainingMap;
415 vector< Vector2d<float> > trainingPixels;
416 vector<string> fields;
418 vector<struct svm_problem> prob(nbag);
419 vector<struct svm_node *> x_space(nbag);
421 for(
int ibag=0;ibag<nbag;++ibag){
423 if(ibag<training_opt.size()){
425 trainingPixels.clear();
426 if(verbose_opt[0]>=1)
427 std::cout <<
"reading imageVector file " << training_opt[0] << std::endl;
431 totalSamples=trainingReaderBag.readDataImageOgr(trainingMap,fields,band_opt,label_opt[0],tlayer_opt,verbose_opt[0]);
433 totalSamples=trainingReaderBag.readDataImageOgr(trainingMap,fields,0,0,label_opt[0],tlayer_opt,verbose_opt[0]);
434 if(trainingMap.size()<2){
435 string errorstring=
"Error: could not read at least two classes from training file, did you provide class labels in training sample (see option label)?";
438 trainingReaderBag.close();
441 cerr << error << std::endl;
444 catch(std::exception& e){
445 std::cerr <<
"Error: ";
446 std::cerr << e.what() << std::endl;
447 std::cerr << CPLGetLastErrorMsg() << std::endl;
451 cerr <<
"error catched" << std::endl;
458 std::cout <<
"training pixels: " << std::endl;
459 map<string,Vector2d<float> >::iterator mapit=trainingMap.begin();
460 while(mapit!=trainingMap.end()){
462 if((mapit->second).size()<minSize_opt[0]){
463 trainingMap.erase(mapit);
466 trainingPixels.push_back(mapit->second);
468 std::cout << mapit->first <<
": " << (mapit->second).size() <<
" samples" << std::endl;
472 nclass=trainingPixels.size();
473 if(classname_opt.size())
474 assert(nclass==classname_opt.size());
475 nband=trainingPixels[0][0].size()-2;
478 assert(nclass==trainingPixels.size());
479 assert(nband==trainingPixels[0][0].size()-2);
484 if(balance_opt[0]>0){
485 while(balance_opt.size()<nclass)
486 balance_opt.push_back(balance_opt.back());
490 for(
short iclass=0;iclass<nclass;++iclass){
491 if(trainingPixels[iclass].size()>balance_opt[iclass]){
492 while(trainingPixels[iclass].size()>balance_opt[iclass]){
493 int index=rand()%trainingPixels[iclass].size();
494 trainingPixels[iclass].erase(trainingPixels[iclass].begin()+index);
498 int oldsize=trainingPixels[iclass].size();
499 for(
int isample=trainingPixels[iclass].size();isample<balance_opt[iclass];++isample){
500 int index = rand()%oldsize;
501 trainingPixels[iclass].push_back(trainingPixels[iclass][index]);
504 totalSamples+=trainingPixels[iclass].size();
509 offset[ibag].resize(nband);
510 scale[ibag].resize(nband);
511 if(offset_opt.size()>1)
512 assert(offset_opt.size()==nband);
513 if(scale_opt.size()>1)
514 assert(scale_opt.size()==nband);
515 for(
int iband=0;iband<nband;++iband){
516 if(verbose_opt[0]>=1)
517 std::cout <<
"scaling for band" << iband << std::endl;
518 offset[ibag][iband]=(offset_opt.size()==1)?offset_opt[0]:offset_opt[iband];
519 scale[ibag][iband]=(scale_opt.size()==1)?scale_opt[0]:scale_opt[iband];
521 if(scale[ibag][iband]<=0){
522 float theMin=trainingPixels[0][0][iband+startBand];
523 float theMax=trainingPixels[0][0][iband+startBand];
524 for(
short iclass=0;iclass<nclass;++iclass){
525 for(
int isample=0;isample<trainingPixels[iclass].size();++isample){
526 if(theMin>trainingPixels[iclass][isample][iband+startBand])
527 theMin=trainingPixels[iclass][isample][iband+startBand];
528 if(theMax<trainingPixels[iclass][isample][iband+startBand])
529 theMax=trainingPixels[iclass][isample][iband+startBand];
532 offset[ibag][iband]=theMin+(theMax-theMin)/2.0;
533 scale[ibag][iband]=(theMax-theMin)/2.0;
534 if(verbose_opt[0]>=1){
535 std::cout <<
"Extreme image values for band " << iband <<
": [" << theMin <<
"," << theMax <<
"]" << std::endl;
536 std::cout <<
"Using offset, scale: " << offset[ibag][iband] <<
", " << scale[ibag][iband] << std::endl;
537 std::cout <<
"scaled values for band " << iband <<
": [" << (theMin-offset[ibag][iband])/scale[ibag][iband] <<
"," << (theMax-offset[ibag][iband])/scale[ibag][iband] <<
"]" << std::endl;
543 offset[ibag].resize(nband);
544 scale[ibag].resize(nband);
545 for(
int iband=0;iband<nband;++iband){
546 offset[ibag][iband]=offset[0][iband];
547 scale[ibag][iband]=scale[0][iband];
552 if(priors_opt.size()==1){
553 priors.resize(nclass);
554 for(
short iclass=0;iclass<nclass;++iclass)
555 priors[iclass]=1.0/nclass;
557 assert(priors_opt.size()==1||priors_opt.size()==nclass);
560 while(bagSize_opt.size()<nclass)
561 bagSize_opt.push_back(bagSize_opt.back());
563 if(verbose_opt[0]>=1){
564 std::cout <<
"number of bands: " << nband << std::endl;
565 std::cout <<
"number of classes: " << nclass << std::endl;
566 if(priorimg_opt.empty()){
567 std::cout <<
"priors:";
568 for(
short iclass=0;iclass<nclass;++iclass)
569 std::cout <<
" " << priors[iclass];
570 std::cout << std::endl;
573 map<string,Vector2d<float> >::iterator mapit=trainingMap.begin();
576 while(mapit!=trainingMap.end()){
577 nameVector.push_back(mapit->first);
578 if(classValueMap.size()){
580 if(classValueMap[mapit->first]>0){
581 if(cm.getClassIndex(type2string<short>(classValueMap[mapit->first]))<0){
582 cm.pushBackClassName(type2string<short>(classValueMap[mapit->first]),doSort);
586 std::cerr <<
"Error: names in classname option are not complete, please check names in training vector and make sure classvalue is > 0" << std::endl;
591 cm.pushBackClassName(mapit->first,doSort);
596 std::cerr <<
"Error: did you provide class pairs names (-c) and integer values (-r) for each class in training vector?" << std::endl;
599 if(classname_opt.empty()){
601 for(
int iclass=0;iclass<nclass;++iclass){
603 std::cout << iclass <<
" " << cm.getClass(iclass) <<
" -> " << string2type<short>(cm.getClass(iclass)) << std::endl;
604 classValueMap[cm.getClass(iclass)]=string2type<short>(cm.getClass(iclass));
616 vector< Vector2d<float> > trainingFeatures(nclass);
617 for(
short iclass=0;iclass<nclass;++iclass){
619 if(verbose_opt[0]>=1)
620 std::cout <<
"calculating features for class " << iclass << std::endl;
623 nctraining=(bagSize_opt[iclass]<100)? trainingPixels[iclass].size()/100.0*bagSize_opt[iclass] : trainingPixels[iclass].size();
626 assert(nctraining<=trainingPixels[iclass].size());
628 if(bagSize_opt[iclass]<100)
629 random_shuffle(trainingPixels[iclass].begin(),trainingPixels[iclass].end());
631 std::cout <<
"nctraining (class " << iclass <<
"): " << nctraining << std::endl;
632 trainingFeatures[iclass].resize(nctraining);
633 for(
int isample=0;isample<nctraining;++isample){
635 for(
int iband=0;iband<nband;++iband){
636 float value=trainingPixels[iclass][isample][iband+startBand];
637 trainingFeatures[iclass][isample].push_back((value-offset[ibag][iband])/scale[ibag][iband]);
640 assert(trainingFeatures[iclass].size()==nctraining);
643 unsigned int nFeatures=trainingFeatures[0][0].size();
644 if(verbose_opt[0]>=1)
645 std::cout <<
"number of features: " << nFeatures << std::endl;
646 unsigned int ntraining=0;
647 for(
short iclass=0;iclass<nclass;++iclass)
648 ntraining+=trainingFeatures[iclass].size();
649 if(verbose_opt[0]>=1)
650 std::cout <<
"training size over all classes: " << ntraining << std::endl;
652 prob[ibag].l=ntraining;
653 prob[ibag].y = Malloc(
double,prob[ibag].l);
654 prob[ibag].x = Malloc(
struct svm_node *,prob[ibag].l);
655 x_space[ibag] = Malloc(
struct svm_node,(nFeatures+1)*ntraining);
656 unsigned long int spaceIndex=0;
658 for(
short iclass=0;iclass<nclass;++iclass){
659 for(
int isample=0;isample<trainingFeatures[iclass].size();++isample){
660 prob[ibag].x[lIndex]=&(x_space[ibag][spaceIndex]);
661 for(
int ifeature=0;ifeature<nFeatures;++ifeature){
662 x_space[ibag][spaceIndex].index=ifeature+1;
663 x_space[ibag][spaceIndex].value=trainingFeatures[iclass][isample][ifeature];
666 x_space[ibag][spaceIndex++].index=-1;
667 prob[ibag].y[lIndex]=iclass;
671 assert(lIndex==prob[ibag].l);
674 param[ibag].svm_type = svmMap[svm_type_opt[0]];
675 param[ibag].kernel_type = kernelMap[kernel_type_opt[0]];
676 param[ibag].degree = kernel_degree_opt[0];
677 param[ibag].gamma = (gamma_opt[0]>0)? gamma_opt[0] : 1.0/nFeatures;
678 param[ibag].coef0 = coef0_opt[0];
679 param[ibag].nu = nu_opt[0];
680 param[ibag].cache_size = cache_opt[0];
681 param[ibag].C = ccost_opt[0];
682 param[ibag].eps = epsilon_tol_opt[0];
683 param[ibag].p = epsilon_loss_opt[0];
684 param[ibag].shrinking = (shrinking_opt[0])? 1 : 0;
685 param[ibag].probability = (prob_est_opt[0])? 1 : 0;
686 param[ibag].nr_weight = 0;
687 param[ibag].weight_label = NULL;
688 param[ibag].weight = NULL;
689 param[ibag].verbose=(verbose_opt[0]>1)?
true:
false;
692 std::cout <<
"checking parameters" << std::endl;
693 svm_check_parameter(&prob[ibag],¶m[ibag]);
695 std::cout <<
"parameters ok, training" << std::endl;
696 svm[ibag]=svm_train(&prob[ibag],¶m[ibag]);
698 std::cout <<
"SVM is now trained" << std::endl;
701 std::cout <<
"Cross validating" << std::endl;
702 double *target = Malloc(
double,prob[ibag].l);
703 svm_cross_validation(&prob[ibag],¶m[ibag],cv_opt[0],target);
704 assert(param[ibag].svm_type != EPSILON_SVR&¶m[ibag].svm_type != NU_SVR);
706 for(
int i=0;i<prob[ibag].l;i++){
707 string refClassName=nameVector[prob[ibag].y[i]];
708 string className=nameVector[target[i]];
709 if(classValueMap.size())
710 cm.incrementResult(type2string<short>(classValueMap[refClassName]),type2string<short>(classValueMap[className]),1.0/nbag);
712 cm.incrementResult(cm.getClass(prob[ibag].y[i]),cm.getClass(target[i]),1.0/nbag);
721 assert(cm.nReference());
722 cm.setFormat(cmformat_opt[0]);
723 cm.reportSE95(
false);
724 std::cout << cm << std::endl;
743 if(input_opt.empty())
746 const char* pszMessage;
747 void* pProgressArg=NULL;
748 GDALProgressFunc pfnProgress=GDALTermProgress;
751 pfnProgress(progress,pszMessage,pProgressArg);
753 bool inputIsRaster=
false;
756 imgReaderOgr.open(input_opt[0]);
757 imgReaderOgr.close();
759 catch(
string errorString){
765 if(verbose_opt[0]>=1)
766 std::cout <<
"opening image " << input_opt[0] << std::endl;
767 testImage.open(input_opt[0]);
770 cerr << error << std::endl;
774 if(priorimg_opt.size()){
776 if(verbose_opt[0]>=1)
777 std::cout <<
"opening prior image " << priorimg_opt[0] << std::endl;
778 priorReader.open(priorimg_opt[0]);
779 assert(priorReader.nrOfCol()==testImage.nrOfCol());
780 assert(priorReader.nrOfRow()==testImage.nrOfRow());
783 cerr << error << std::endl;
787 cerr <<
"error catched" << std::endl;
792 int nrow=testImage.nrOfRow();
793 int ncol=testImage.nrOfCol();
794 if(option_opt.findSubstring(
"INTERLEAVE=")==option_opt.end()){
795 string theInterleave=
"INTERLEAVE=";
796 theInterleave+=testImage.getInterleave();
797 option_opt.push_back(theInterleave);
799 vector<char> classOut(ncol);
807 string imageType=testImage.getImageType();
808 if(oformat_opt.size())
809 imageType=oformat_opt[0];
811 assert(output_opt.size());
812 if(verbose_opt[0]>=1)
813 std::cout <<
"opening class image for writing output " << output_opt[0] << std::endl;
814 if(classBag_opt.size()){
815 classImageBag.open(classBag_opt[0],ncol,nrow,nbag,GDT_Byte,imageType,option_opt);
816 classImageBag.GDALSetNoDataValue(nodata_opt[0]);
817 classImageBag.copyGeoTransform(testImage);
818 classImageBag.setProjection(testImage.getProjection());
820 classImageOut.open(output_opt[0],ncol,nrow,1,GDT_Byte,imageType,option_opt);
821 classImageOut.GDALSetNoDataValue(nodata_opt[0]);
822 classImageOut.copyGeoTransform(testImage);
823 classImageOut.setProjection(testImage.getProjection());
824 if(colorTable_opt.size())
825 classImageOut.setColorTable(colorTable_opt[0],0);
827 probImage.open(prob_opt[0],ncol,nrow,nclass,GDT_Byte,imageType,option_opt);
828 probImage.GDALSetNoDataValue(nodata_opt[0]);
829 probImage.copyGeoTransform(testImage);
830 probImage.setProjection(testImage.getProjection());
832 if(entropy_opt.size()){
833 entropyImage.open(entropy_opt[0],ncol,nrow,1,GDT_Byte,imageType,option_opt);
834 entropyImage.GDALSetNoDataValue(nodata_opt[0]);
835 entropyImage.copyGeoTransform(testImage);
836 entropyImage.setProjection(testImage.getProjection());
840 cerr << error << std::endl;
847 maskWriter.open(
"/vsimem/mask.tif",ncol,nrow,1,GDT_Float32,imageType,option_opt);
848 maskWriter.GDALSetNoDataValue(nodata_opt[0]);
849 maskWriter.copyGeoTransform(testImage);
850 maskWriter.setProjection(testImage.getProjection());
851 vector<double> burnValues(1,1);
852 maskWriter.rasterizeOgr(extentReader,burnValues);
853 extentReader.close();
857 cerr << error << std::endl;
861 cerr <<
"error catched" << std::endl;
865 mask_opt.push_back(
"/vsimem/mask.tif");
870 if(verbose_opt[0]>=1)
871 std::cout <<
"opening mask image file " << mask_opt[0] << std::endl;
872 maskReader.open(mask_opt[0]);
875 cerr << error << std::endl;
879 cerr <<
"error catched" << std::endl;
884 for(
int iline=0;iline<nrow;++iline){
885 vector<float> buffer(ncol);
886 vector<short> lineMask;
888 if(priorimg_opt.size())
889 linePrior.resize(nclass,ncol);
892 vector<float> entropy(ncol);
894 if(classBag_opt.size())
895 classBag.resize(nbag,ncol);
898 for(
int iband=0;iband<band_opt.size();++iband){
899 if(verbose_opt[0]==2)
900 std::cout <<
"reading band " << band_opt[iband] << std::endl;
901 assert(band_opt[iband]>=0);
902 assert(band_opt[iband]<testImage.nrOfBand());
903 testImage.readData(buffer,GDT_Float32,iline,band_opt[iband]);
904 for(
int icol=0;icol<ncol;++icol)
905 hpixel[icol].push_back(buffer[icol]);
909 for(
int iband=0;iband<nband;++iband){
910 if(verbose_opt[0]==2)
911 std::cout <<
"reading band " << iband << std::endl;
913 assert(iband<testImage.nrOfBand());
914 testImage.readData(buffer,GDT_Float32,iline,iband);
915 for(
int icol=0;icol<ncol;++icol)
916 hpixel[icol].push_back(buffer[icol]);
920 catch(
string theError){
921 cerr <<
"Error reading " << input_opt[0] <<
": " << theError << std::endl;
925 cerr <<
"error catched" << std::endl;
928 assert(nband==hpixel[0].size());
930 std::cout <<
"used bands: " << nband << std::endl;
932 if(priorimg_opt.size()){
934 for(
short iclass=0;iclass<nclass;++iclass){
935 if(verbose_opt.size()>1)
936 std::cout <<
"Reading " << priorimg_opt[0] <<
" band " << iclass <<
" line " << iline << std::endl;
937 priorReader.readData(linePrior[iclass],GDT_Float32,iline,iclass);
940 catch(
string theError){
941 std::cerr <<
"Error reading " << priorimg_opt[0] <<
": " << theError << std::endl;
945 cerr <<
"error catched" << std::endl;
949 double oldRowMask=-1;
951 for(
int icol=0;icol<ncol;++icol){
952 assert(hpixel[icol].size()==nband);
953 bool doClassify=
true;
959 testImage.image2geo(icol,iline,geox,geoy);
961 if(uly>=geoy&&lry<=geoy&&ulx<=geox&&lrx>=geox){
970 testImage.image2geo(icol,iline,geox,geoy);
971 maskReader.geo2image(geox,geoy,colMask,rowMask);
972 colMask=
static_cast<int>(colMask);
973 rowMask=
static_cast<int>(rowMask);
974 if(rowMask>=0&&rowMask<maskReader.nrOfRow()&&colMask>=0&&colMask<maskReader.nrOfCol()){
975 if(static_cast<int>(rowMask)!=
static_cast<int>(oldRowMask)){
976 assert(rowMask>=0&&rowMask<maskReader.nrOfRow());
979 maskReader.readData(lineMask,GDT_Int16,static_cast<int>(rowMask));
981 catch(
string errorstring){
982 cerr << errorstring << endl;
986 cerr <<
"error catched" << std::endl;
992 for(
short ivalue=0;ivalue<msknodata_opt.size();++ivalue){
993 if(msknodata_opt[ivalue]>=0){
994 if(lineMask[colMask]==msknodata_opt[ivalue]){
995 theMask=lineMask[colMask];
1001 if(lineMask[colMask]!=-msknodata_opt[ivalue]){
1002 theMask=lineMask[colMask];
1012 if(classBag_opt.size())
1013 for(
int ibag=0;ibag<nbag;++ibag)
1014 classBag[ibag][icol]=theMask;
1015 classOut[icol]=theMask;
1020 for(
int iband=0;iband<hpixel[icol].size();++iband){
1021 if(hpixel[icol][iband]){
1029 for(
short iclass=0;iclass<nclass;++iclass)
1030 probOut[iclass][icol]=0;
1032 if(classBag_opt.size())
1033 for(
int ibag=0;ibag<nbag;++ibag)
1034 classBag[ibag][icol]=nodata_opt[0];
1035 classOut[icol]=nodata_opt[0];
1038 if(verbose_opt[0]>1)
1039 std::cout <<
"begin classification " << std::endl;
1041 for(
int ibag=0;ibag<nbag;++ibag){
1042 vector<double> result(nclass);
1045 for(
int iband=0;iband<nband;++iband){
1046 x[iband].index=iband+1;
1047 x[iband].value=(hpixel[icol][iband]-offset[ibag][iband])/scale[ibag][iband];
1050 double predict_label=0;
1051 vector<float> prValues(nclass);
1053 if(!prob_est_opt[0]){
1054 predict_label = svm_predict(svm[ibag],x);
1055 for(
short iclass=0;iclass<nclass;++iclass){
1056 if(iclass==static_cast<short>(predict_label))
1063 assert(svm_check_probability_model(svm[ibag]));
1064 predict_label = svm_predict_probability(svm[ibag],x,&(result[0]));
1067 if(classBag_opt.size()){
1070 classBag[ibag][icol]=0;
1073 if(priorimg_opt.size()){
1074 for(
short iclass=0;iclass<nclass;++iclass)
1075 normPrior+=linePrior[iclass][icol];
1077 for(
short iclass=0;iclass<nclass;++iclass){
1078 if(priorimg_opt.size())
1079 priors[iclass]=linePrior[iclass][icol]/normPrior;
1080 switch(comb_opt[0]){
1083 probOut[iclass][icol]+=result[iclass]*priors[iclass];
1086 probOut[iclass][icol]*=pow(static_cast<float>(priors[iclass]),static_cast<float>(1.0-nbag)/nbag)*result[iclass];
1089 if(priors[iclass]*result[iclass]>probOut[iclass][icol])
1090 probOut[iclass][icol]=priors[iclass]*result[iclass];
1093 if(classBag_opt.size()){
1099 if(result[iclass]>maxP){
1100 maxP=result[iclass];
1101 classBag[ibag][icol]=iclass;
1112 for(
short iclass=0;iclass<nclass;++iclass){
1113 if(probOut[iclass][icol]>maxBag1){
1114 maxBag1=probOut[iclass][icol];
1115 classOut[icol]=classValueMap[nameVector[iclass]];
1117 else if(probOut[iclass][icol]>maxBag2)
1118 maxBag2=probOut[iclass][icol];
1119 normBag+=probOut[iclass][icol];
1123 for(
short iclass=0;iclass<nclass;++iclass){
1124 float prv=probOut[iclass][icol];
1126 entropy[icol]-=prv*log(prv)/log(2.0);
1129 probOut[iclass][icol]=
static_cast<short>(prv+0.5);
1134 entropy[icol]/=log(static_cast<double>(nclass))/log(2.0);
1135 entropy[icol]=
static_cast<short>(100*entropy[icol]+0.5);
1136 if(active_opt.size()){
1137 if(entropy[icol]>activePoints.back().value){
1138 activePoints.back().value=entropy[icol];
1139 activePoints.back().posx=icol;
1140 activePoints.back().posy=iline;
1143 std::cout << activePoints.back().posx <<
" " << activePoints.back().posy <<
" " << activePoints.back().value << std::endl;
1148 if(classBag_opt.size())
1149 for(
int ibag=0;ibag<nbag;++ibag)
1150 classImageBag.writeData(classBag[ibag],GDT_Byte,iline,ibag);
1151 if(prob_opt.size()){
1152 for(
short iclass=0;iclass<nclass;++iclass)
1153 probImage.writeData(probOut[iclass],GDT_Float32,iline,iclass);
1155 if(entropy_opt.size()){
1156 entropyImage.writeData(entropy,GDT_Float32,iline);
1158 classImageOut.writeData(classOut,GDT_Byte,iline);
1159 if(!verbose_opt[0]){
1160 progress=
static_cast<float>(iline+1.0)/classImageOut.nrOfRow();
1161 pfnProgress(progress,pszMessage,pProgressArg);
1165 if(active_opt.size()){
1166 for(
int iactive=0;iactive<activePoints.size();++iactive){
1167 std::map<string,double> pointMap;
1168 for(
int iband=0;iband<testImage.nrOfBand();++iband){
1170 testImage.readData(value,GDT_Float64,static_cast<int>(activePoints[iactive].posx),static_cast<int>(activePoints[iactive].posy),iband);
1173 pointMap[fs.str()]=value;
1175 pointMap[label_opt[0]]=0;
1177 testImage.image2geo(activePoints[iactive].posx,activePoints[iactive].posy,x,y);
1178 std::string fieldname=
"id";
1179 activeWriter.addPoint(x,y,pointMap,fieldname,++nactive);
1186 if(priorimg_opt.size())
1187 priorReader.close();
1190 if(entropy_opt.size())
1191 entropyImage.close();
1192 if(classBag_opt.size())
1193 classImageBag.close();
1194 classImageOut.close();
1199 for(
int ivalidation=0;ivalidation<input_opt.size();++ivalidation){
1200 if(output_opt.size())
1201 assert(output_opt.size()==input_opt.size());
1203 std::cout <<
"opening img reader " << input_opt[ivalidation] << std::endl;
1204 imgReaderOgr.open(input_opt[ivalidation]);
1207 if(output_opt.size()){
1209 std::cout <<
"opening img writer and copying fields from img reader" << output_opt[ivalidation] << std::endl;
1210 imgWriterOgr.open(output_opt[ivalidation],imgReaderOgr);
1213 cout <<
"number of layers in input ogr file: " << imgReaderOgr.getLayerCount() << endl;
1214 for(
int ilayer=0;ilayer<imgReaderOgr.getLayerCount();++ilayer){
1216 cout <<
"processing input layer " << ilayer << endl;
1217 if(output_opt.size()){
1219 std::cout <<
"creating field class" << std::endl;
1220 if(classValueMap.size())
1221 imgWriterOgr.createField(
"class",OFTInteger,ilayer);
1223 imgWriterOgr.createField(
"class",OFTString,ilayer);
1225 unsigned int nFeatures=imgReaderOgr.getFeatureCount(ilayer);
1226 unsigned int ifeature=0;
1228 pfnProgress(progress,pszMessage,pProgressArg);
1229 OGRFeature *poFeature;
1230 while( (poFeature = imgReaderOgr.getLayer(ilayer)->GetNextFeature()) != NULL ){
1231 if(verbose_opt[0]>1)
1232 std::cout <<
"feature " << ifeature << std::endl;
1233 if( poFeature == NULL ){
1234 cout <<
"Warning: could not read feature " << ifeature <<
" in layer " << imgReaderOgr.getLayerName(ilayer) << endl;
1237 OGRFeature *poDstFeature = NULL;
1238 if(output_opt.size()){
1239 poDstFeature=imgWriterOgr.createFeature(ilayer);
1240 if( poDstFeature->SetFrom( poFeature, TRUE ) != OGRERR_NONE ){
1241 CPLError( CE_Failure, CPLE_AppDefined,
1242 "Unable to translate feature %d from layer %s.\n",
1243 poFeature->GetFID(), imgWriterOgr.getLayerName(ilayer).c_str() );
1244 OGRFeature::DestroyFeature( poFeature );
1245 OGRFeature::DestroyFeature( poDstFeature );
1248 vector<float> validationPixel;
1249 vector<float> validationFeature;
1251 imgReaderOgr.readData(validationPixel,OFTReal,fields,poFeature,ilayer);
1252 assert(validationPixel.size()==nband);
1253 vector<float> probOut(nclass);
1254 for(
short iclass=0;iclass<nclass;++iclass)
1256 for(
int ibag=0;ibag<nbag;++ibag){
1257 for(
int iband=0;iband<nband;++iband){
1258 validationFeature.push_back((validationPixel[iband]-offset[ibag][iband])/scale[ibag][iband]);
1259 if(verbose_opt[0]==2)
1260 std::cout <<
" " << validationFeature.back();
1262 if(verbose_opt[0]==2)
1263 std::cout << std::endl;
1264 vector<double> result(nclass);
1266 x = (
struct svm_node *) malloc((validationFeature.size()+1)*
sizeof(
struct svm_node));
1267 for(
int i=0;i<validationFeature.size();++i){
1269 x[i].value=validationFeature[i];
1272 x[validationFeature.size()].index=-1;
1273 double predict_label=0;
1274 if(!prob_est_opt[0]){
1275 predict_label = svm_predict(svm[ibag],x);
1276 for(
short iclass=0;iclass<nclass;++iclass){
1277 if(iclass==static_cast<short>(predict_label))
1284 assert(svm_check_probability_model(svm[ibag]));
1285 predict_label = svm_predict_probability(svm[ibag],x,&(result[0]));
1287 if(verbose_opt[0]>1){
1288 std::cout <<
"predict_label: " << predict_label << std::endl;
1289 for(
int iclass=0;iclass<result.size();++iclass)
1290 std::cout << result[iclass] <<
" ";
1291 std::cout << std::endl;
1295 for(
short iclass=0;iclass<nclass;++iclass){
1296 switch(comb_opt[0]){
1299 probOut[iclass]+=result[iclass]*priors[iclass];
1302 probOut[iclass]*=pow(static_cast<float>(priors[iclass]),static_cast<float>(1.0-nbag)/nbag)*result[iclass];
1305 if(priors[iclass]*result[iclass]>probOut[iclass])
1306 probOut[iclass]=priors[iclass]*result[iclass];
1316 string classOut=
"Unclassified";
1317 for(
short iclass=0;iclass<nclass;++iclass){
1318 if(verbose_opt[0]>1)
1319 std::cout << probOut[iclass] <<
" ";
1320 if(probOut[iclass]>maxBag){
1321 maxBag=probOut[iclass];
1322 classOut=nameVector[iclass];
1326 if(verbose_opt[0]>1){
1327 if(classValueMap.size())
1328 std::cout <<
"->" << classValueMap[classOut] << std::endl;
1330 std::cout <<
"->" << classOut << std::endl;
1332 if(output_opt.size()){
1333 if(classValueMap.size())
1334 poDstFeature->SetField(
"class",classValueMap[classOut]);
1336 poDstFeature->SetField(
"class",classOut.c_str());
1337 poDstFeature->SetFID( poFeature->GetFID() );
1339 int labelIndex=poFeature->GetFieldIndex(label_opt[0].c_str());
1341 string classRef=poFeature->GetFieldAsString(labelIndex);
1343 if(classValueMap.size())
1344 cm.incrementResult(type2string<short>(classValueMap[classRef]),type2string<short>(classValueMap[classOut]),1);
1346 cm.incrementResult(classRef,classOut,1);
1350 if(output_opt.size()){
1351 if(imgWriterOgr.createFeature(poDstFeature,ilayer) != OGRERR_NONE){
1352 CPLError( CE_Failure, CPLE_AppDefined,
1353 "Unable to translate feature %d from layer %s.\n",
1354 poFeature->GetFID(), imgWriterOgr.getLayerName(ilayer).c_str() );
1355 OGRFeature::DestroyFeature( poDstFeature );
1356 OGRFeature::DestroyFeature( poDstFeature );
1360 if(!verbose_opt[0]){
1361 progress=
static_cast<float>(ifeature+1.0)/nFeatures;
1362 pfnProgress(progress,pszMessage,pProgressArg);
1364 OGRFeature::DestroyFeature( poFeature );
1365 OGRFeature::DestroyFeature( poDstFeature );
1368 imgReaderOgr.close();
1369 if(output_opt.size())
1370 imgWriterOgr.close();
1372 if(cm.nReference()){
1373 std::cout << cm << std::endl;
1374 cout <<
"class #samples userAcc prodAcc" << endl;
1381 for(
short iclass=0;iclass<cm.nClasses();++iclass){
1382 dua=cm.ua_pct(cm.getClass(iclass),&se95_ua);
1383 dpa=cm.pa_pct(cm.getClass(iclass),&se95_pa);
1384 cout << cm.getClass(iclass) <<
" " << cm.nReference(cm.getClass(iclass)) <<
" " << dua <<
" (" << se95_ua <<
")" <<
" " << dpa <<
" (" << se95_pa <<
")" << endl;
1386 std::cout <<
"Kappa: " << cm.kappa() << std::endl;
1387 doa=cm.oa(&se95_oa);
1388 std::cout <<
"Overall Accuracy: " << 100*doa <<
" (" << 100*se95_oa <<
")" << std::endl;
1392 if(active_opt.size())
1393 activeWriter.close();
1395 catch(
string errorString){
1396 std::cerr <<
"Error: errorString" << std::endl;
1399 for(
int ibag=0;ibag<nbag;++ibag){
1401 svm_destroy_param(¶m[ibag]);
1404 free(x_space[ibag]);
1405 svm_free_and_destroy_model(&(svm[ibag]));
throw this class when syntax error in command line option