Clasificación de imágenes con SVM

Librerias aplicadas

library(caret)
library(raster)

Carga de las bandas de imágenes satelitales.

band1 <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/capturas-landsat/roma/LC81910312013208LGN00/LC81910312013208LGN00_B1.TIF")
band2 <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/capturas-landsat/roma/LC81910312013208LGN00/LC81910312013208LGN00_B2.TIF")
band3 <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/capturas-landsat/roma/LC81910312013208LGN00/LC81910312013208LGN00_B3.TIF")
band4 <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/capturas-landsat/roma/LC81910312013208LGN00/LC81910312013208LGN00_B4.TIF")
band5 <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/capturas-landsat/roma/LC81910312013208LGN00/LC81910312013208LGN00_B5.TIF")
band6 <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/capturas-landsat/roma/LC81910312013208LGN00/LC81910312013208LGN00_B6.TIF")
band7 <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/capturas-landsat/roma/LC81910312013208LGN00/LC81910312013208LGN00_B7.TIF")
band10 <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/capturas-landsat/roma/LC81910312013208LGN00/LC81910312013208LGN00_B10.TIF")
band11 <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/capturas-landsat/roma/LC81910312013208LGN00/LC81910312013208LGN00_B11.TIF")

roma_rst_lan <- brick(band1,band2,band3,band4,band5,band6,band7,band10,band11)
names(roma_rst_lan) <- c(paste("Banda",1:7,sep=""),"Banda10","Banda11")

Procesado de datos

#Carga de muestras de entrenamiento
rstTrainroma <- raster("/Users/elianaalizchuquillanquijulcapari/Documents/workspace-r/demos/DataTraining/rome_lcz_GT.tif")

# Remover ceros (fondo NA)
rstTrainroma[rstTrainroma==0] <- NA

# Convertir a tipo discreto
rstTrainroma <- ratify(rstTrainroma)

# Cambiamos el nombre
names(rstTrainroma) <- "layer"

#Muestras de valores
table(values(rstTrainroma))

   2    3    5    6    8   10   11   12   14   17 
1551  104 1495  480  435   51  284  555  984  500 

# Eliminación de NAs 
rstDFlanroma.df <- na.omit(values(stack(rstTrainroma, roma_rst_lan)))

cols <- c("#fde725","#b5de2b", "#6ece58","#35b779", "#1f9e89","#26828e", "#31688e", "#3e4989", "#482878", "#440154")
plot(rstTrainroma, main="Muestras de entrenamiento de Roma", col = cols)

Fase de Clasificación

#Si es la primera vez se ejecuta, sino ya esta guardado los datos del dataframe.
#write.csv(rstDFlanroma.df,"rstDFRoma.csv", row.names = FALSE)

#Leemos los datos
samplesLanroma = read.csv("rstDFRoma.csv")

# Se divide las muestras en 70 y 30
trainx = list(0)
evalx = list(0)
for (i in 1:17){ # se itera en todas las posibles categorías
  cls = samplesLanroma[samplesLanroma$layer == i,]
  smpl <- floor(0.70 * nrow(cls))
  tt <- sample(seq_len(nrow(cls)), size = smpl)
  trainx[[i]] <- cls[tt,]
  evalx[[i]] <- cls[-tt,]
}


# Se combina en dataframe referenciados s entrenamiento y evaluación
trn = do.call(rbind, trainx) 
eva = do.call(rbind, evalx)

Algoritmo SVM

# Definiendo el control del entrenamiento
fitControl2 <- trainControl(method = "cv", # repeated cross-validation of the training data
                   number = 10) 


# Entrenamiento con svm
svm_model1<- caret::train(x = trn[,(2:ncol(trn))], y = as.factor(trn$layer),
                      method = "svmRadial", metric="Accuracy", 
                      trControl = fitControl2, 
                      preProcess = c("center", "scale"),
                      tuneLength = 10)
# Entrenamiento modelo 2
svm_grid <- expand.grid(sigma=seq(from = 0.01, to = 0.10, by = 0.02),
                        C=seq(from = 2, to = 10, by = 2))

svm_model2<- caret::train(x = trn[,(2:ncol(trn))], y = as.factor(trn$layer),
                      method = "svmRadialSigma", metric="Accuracy", 
                      trControl = fitControl2,
                      preProcess = c("center", "scale"),
                      tuneGrid = svm_grid)

Exploración del modelo

#Información modelo svm
print(svm_model1)

Support Vector Machines with Radial Basis Function Kernel 

4502 samples
   9 predictor
  10 classes: '2', '3', '5', '6', '8', '10', '11', '12', '14', '17' 

Pre-processing: centered (9), scaled (9) 
Resampling: Cross-Validated (10 fold) 
Summary of sample sizes: 4051, 4052, 4053, 4053, 4051, 4051, ... 
Resampling results across tuning parameters:

  C       Accuracy   Kappa    
    0.25  0.7587714  0.7092347
    0.50  0.7734297  0.7271917
    1.00  0.7825404  0.7382531
    2.00  0.7945404  0.7528460
    4.00  0.8034259  0.7635539
    8.00  0.8109760  0.7730634
   16.00  0.8187538  0.7828206
   32.00  0.8274191  0.7935093
   64.00  0.8296378  0.7962199
  128.00  0.8247529  0.7903756

Tuning parameter 'sigma' was held constant at a value of 0.3604227
Accuracy was used to select the optimal model using the largest value.
The final values used for the model were sigma = 0.3604227 and C = 64.

print(svm_model2) #modelo 2

Support Vector Machines with Radial Basis Function Kernel 

4502 samples
   9 predictor
  10 classes: '2', '3', '5', '6', '8', '10', '11', '12', '14', '17' 

Pre-processing: centered (9), scaled (9) 
Resampling: Cross-Validated (10 fold) 
Summary of sample sizes: 4051, 4051, 4050, 4054, 4053, 4052, ... 
Resampling results across tuning parameters:

  sigma  C   Accuracy   Kappa    
  0.01    2  0.7425466  0.6887106
  0.01    4  0.7536667  0.7026741
  0.01    6  0.7581023  0.7081204
  0.01    8  0.7583260  0.7084588
  0.01   10  0.7616618  0.7125652
  0.03    2  0.7585497  0.7087512
  0.03    4  0.7645433  0.7162499
  0.03    6  0.7729819  0.7265514
  0.03    8  0.7798758  0.7349210
  0.03   10  0.7834408  0.7393069
  0.05    2  0.7638771  0.7154095
  0.05    4  0.7789864  0.7338853
  0.05    6  0.7845485  0.7407388
  0.05    8  0.7912068  0.7487595
  0.05   10  0.7960883  0.7546275
  0.07    2  0.7761000  0.7303001
  0.07    4  0.7843252  0.7405191
  0.07    6  0.7925387  0.7503611
  0.07    8  0.7967565  0.7555402
  0.07   10  0.8003101  0.7598150
  0.09    2  0.7798852  0.7350146
  0.09    4  0.7889885  0.7461249
  0.09    6  0.7956468  0.7542104
  0.09    8  0.7989767  0.7582768
  0.09   10  0.8000943  0.7596269

Accuracy was used to select the optimal model using the largest value.
The final values used for the model were sigma = 0.07 and C = 10.

#Gráfica del modelo svm
plot(svm_model1)

plot(svm_model2) #modelo 2

Evaluación del modelo 1.

pred_svm_model1<- predict(svm_model1,newdata = eva[,-1])

cm <- confusionMatrix(data = pred_svm_model1, as.factor(eva$layer))
cm

Confusion Matrix and Statistics

          Reference
Prediction   2   3   5   6   8  10  11  12  14  17
        2  398   6  72   1   3   2   0   0   0   0
        3    5  24   3   1   0   0   0   0   0   0
        5   58   2 345  20  11   7   1   4   1   0
        6    4   0  11 107   0   1   1   5   4   0
        8    1   0   7   0 117   0   0   0   0   0
        10   0   0   0   0   0   5   0   0   0   0
        11   0   0   1   0   0   0  83   3   0   0
        12   0   0   8  10   0   0   1 150   2   0
        14   0   0   2   5   0   1   0   5 289   0
        17   0   0   0   0   0   0   0   0   0 150

Overall Statistics
                                          
               Accuracy : 0.8611          
                 95% CI : (0.8449, 0.8762)
    No Information Rate : 0.2406          
    P-Value [Acc > NIR] : < 2.2e-16       
                                          
                  Kappa : 0.8341          
                                          
 Mcnemar's Test P-Value : NA              

Statistics by Class:

                     Class: 2 Class: 3 Class: 5 Class: 6 Class: 8 Class: 10
Sensitivity            0.8541  0.75000   0.7684  0.74306  0.89313  0.312500
Specificity            0.9429  0.99528   0.9301  0.98550  0.99557  1.000000
Pos Pred Value         0.8257  0.72727   0.7684  0.80451  0.93600  1.000000
Neg Pred Value         0.9533  0.99580   0.9301  0.97949  0.99227  0.994306
Prevalence             0.2406  0.01652   0.2318  0.07434  0.06763  0.008260
Detection Rate         0.2055  0.01239   0.1781  0.05524  0.06040  0.002581
Detection Prevalence   0.2488  0.01704   0.2318  0.06866  0.06453  0.002581
Balanced Accuracy      0.8985  0.87264   0.8492  0.86428  0.94435  0.656250
                     Class: 11 Class: 12 Class: 14 Class: 17
Sensitivity            0.96512   0.89820    0.9764   1.00000
Specificity            0.99784   0.98814    0.9921   1.00000
Pos Pred Value         0.95402   0.87719    0.9570   1.00000
Neg Pred Value         0.99838   0.99037    0.9957   1.00000
Prevalence             0.04440   0.08622    0.1528   0.07744
Detection Rate         0.04285   0.07744    0.1492   0.07744
Detection Prevalence   0.04491   0.08828    0.1559   0.07744
Balanced Accuracy      0.98148   0.94317    0.9842   1.00000

Representación de Matriz de confusión

cm_d <- as.data.frame(cm$table)

cm_st <-data.frame(cm$overall)

cm_st$cm.overall <- round(cm_st$cm.overall,2)

cm_p <- as.data.frame(prop.table(cm$table))
cm_d$Perc <- round(cm_p$Freq*100,2)

# dibujando la matriz
ggplot(data = cm_d, aes(x = Prediction , y =  Reference, fill = Freq))+
  geom_tile(color = "white", lwd = 0.5) +
  geom_text(aes(label = paste("",Freq)), color = 'white', size = 3.5)+ 
  theme_minimal() +
  ggtitle("Matriz de Confusión")

Evaluación del modelo 2.

pred_svm_model2<- predict(svm_model2,newdata = eva[,-1])

cm <- confusionMatrix(data = pred_svm_model2, as.factor(eva$layer))
cm

Confusion Matrix and Statistics

          Reference
Prediction   2   3   5   6   8  10  11  12  14  17
        2  377  26  74   2   6   5   0   0   0   0
        3    0   0   0   0   0   0   0   0   0   0
        5   80   4 336  25  19  10   0  18   6   0
        6    5   2  19  98   0   0   0   2   2   0
        8    4   0   7   0 106   0   0   0   0   0
        10   0   0   0   0   0   1   0   0   0   0
        11   0   0   1   0   0   0  82   3   0   0
        12   0   0  10  15   0   0   3 131  12   0
        14   0   0   2   4   0   0   1  13 276   0
        17   0   0   0   0   0   0   0   0   0 150

Overall Statistics
                                          
               Accuracy : 0.8038          
                 95% CI : (0.7854, 0.8213)
    No Information Rate : 0.2406          
    P-Value [Acc > NIR] : < 2.2e-16       
                                          
                  Kappa : 0.7641          
                                          
 Mcnemar's Test P-Value : NA              

Statistics by Class:

                     Class: 2 Class: 3 Class: 5 Class: 6 Class: 8 Class: 10
Sensitivity            0.8090  0.00000   0.7483  0.68056  0.80916 0.0625000
Specificity            0.9232  1.00000   0.8911  0.98327  0.99391 1.0000000
Pos Pred Value         0.7694      NaN   0.6747  0.76562  0.90598 1.0000000
Neg Pred Value         0.9385  0.98348   0.9215  0.97457  0.98626 0.9922521
Prevalence             0.2406  0.01652   0.2318  0.07434  0.06763 0.0082602
Detection Rate         0.1946  0.00000   0.1735  0.05059  0.05472 0.0005163
Detection Prevalence   0.2530  0.00000   0.2571  0.06608  0.06040 0.0005163
Balanced Accuracy      0.8661  0.50000   0.8197  0.83191  0.90153 0.5312500
                     Class: 11 Class: 12 Class: 14 Class: 17
Sensitivity            0.95349   0.78443    0.9324   1.00000
Specificity            0.99784   0.97740    0.9878   1.00000
Pos Pred Value         0.95349   0.76608    0.9324   1.00000
Neg Pred Value         0.99784   0.97961    0.9878   1.00000
Prevalence             0.04440   0.08622    0.1528   0.07744
Detection Rate         0.04233   0.06763    0.1425   0.07744
Detection Prevalence   0.04440   0.08828    0.1528   0.07744
Balanced Accuracy      0.97566   0.88092    0.9601   1.00000

Representación de Matriz de confusión

cm_d <- as.data.frame(cm$table)

cm_st <-data.frame(cm$overall)

cm_st$cm.overall <- round(cm_st$cm.overall,2)

cm_p <- as.data.frame(prop.table(cm$table))
cm_d$Perc <- round(cm_p$Freq*100,2)

# dibujando la matriz
ggplot(data = cm_d, aes(x = Prediction , y =  Reference, fill = Freq))+
  geom_tile(color = "white", lwd = 0.5) +
  geom_text(aes(label = paste("",Freq)), color = 'white', size = 3.5)+ 
  theme_minimal() +
  ggtitle("Matriz de Confusión")

Predicción y visualización del modelo 1.

classi_svm_model1 <- raster::predict(roma_rst_lan, model=svm_model1)
plot(classi_svm_model1, main="Clasificación SVM Modelo 1",col = cols)

Predicción y visualización del modelo 2.

classi_svm_model2 <- raster::predict(roma_rst_lan, model=svm_model2)
plot(classi_svm_model2, main="Clasificación SVM Modelo 2",col = cols)