Clasificación de imágenes con Random Forest

Librerias aplicadas

library(caret)

Loading required package: ggplot2
Loading required package: lattice

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)))

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

NA
NA

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 Random Forest

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


# Entrenamiento con rf
rf_model1<- caret::train(x = trn[,(2:ncol(trn))], y = as.factor(trn$layer),
                        method = "rf", 
                        metric="Accuracy", 
                        trControl = fitControl2,
                        preProcess = c("center", "scale"),
                        tuneLength  = 8)

rf_grid <- expand.grid(mtry=1:20)

rf_model2<- caret::train(x = trn[,(2:ncol(trn))], y = as.factor(trn$layer),
                        method = "rf", 
                        metric="Accuracy", 
                        trControl = fitControl2,
                        preProcess = c("center", "scale"),
                        tuneGrid = rf_grid)

Exploración del modelo

#Información modelo rf
print(rf_model1)

Random Forest 

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: 4050, 4054, 4055, 4052, 4050, 4053, ... 
Resampling results across tuning parameters:

  mtry  Accuracy   Kappa    
  2     0.7756809  0.7309041
  3     0.7750019  0.7302374
  4     0.7763327  0.7319464
  5     0.7774533  0.7332007
  6     0.7778863  0.7338051
  7     0.7765584  0.7322465
  8     0.7745545  0.7298555
  9     0.7752167  0.7306680

Accuracy was used to select the optimal model using the largest value.
The final value used for the model was mtry = 6.

print(rf_model1$finalModel)

Call:
 randomForest(x = x, y = y, mtry = min(param$mtry, ncol(x))) 
               Type of random forest: classification
                     Number of trees: 500
No. of variables tried at each split: 6

        OOB estimate of  error rate: 22.61%
Confusion matrix:
     2  3   5   6   8 10  11  12  14  17 class.error
2  856 11 194   8  13  1   0   0   2   0  0.21105991
3   34 18  16   4   0  0   0   0   0   0  0.75000000
5  213  4 706  53  24  4   1  20  21   0  0.32504780
6   10  2  62 198   0  1   0  44  19   0  0.41071429
8   12  0  40   1 249  0   0   0   2   0  0.18092105
10  11  0  10   3   0 11   0   0   0   0  0.68571429
11   0  0   1   0   0  0 192   4   1   0  0.03030303
12   1  0  37  21   0  1   5 286  37   0  0.26288660
14   0  0  24  16   2  0   0  28 618   0  0.10174419
17   0  0   0   0   0  0   0   0   0 350  0.00000000

#Gráfica del modelo rf
plot(rf_model1)

print(rf_model2) #modelo 2

Random Forest 

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: 4052, 4053, 4050, 4049, 4052, 4050, ... 
Resampling results across tuning parameters:

  mtry  Accuracy   Kappa    
   1    0.7745519  0.7292963
   2    0.7752407  0.7303372
   3    0.7796803  0.7358375
   4    0.7763415  0.7319764
   5    0.7767909  0.7324974
   6    0.7776748  0.7336103
   7    0.7750096  0.7304041
   8    0.7752426  0.7306268
   9    0.7705848  0.7251480
  10    0.7776926  0.7336253
  11    0.7727987  0.7278105
  12    0.7732461  0.7282577
  13    0.7736925  0.7288293
  14    0.7770155  0.7327538
  15    0.7721334  0.7270123
  16    0.7725750  0.7275754
  17    0.7739039  0.7291407
  18    0.7750170  0.7304012
  19    0.7743503  0.7296546
  20    0.7763557  0.7320517

Accuracy was used to select the optimal model using the largest value.
The final value used for the model was mtry = 3.

print(rf_model2$finalModel)

Call:
 randomForest(x = x, y = y, mtry = min(param$mtry, ncol(x))) 
               Type of random forest: classification
                     Number of trees: 500
No. of variables tried at each split: 3

        OOB estimate of  error rate: 21.83%
Confusion matrix:
     2  3   5   6   8 10  11  12  14  17 class.error
2  864 10 183  11  14  1   0   0   2   0  0.20368664
3   39 18  12   3   0  0   0   0   0   0  0.75000000
5  205  3 722  51  22  4   1  15  23   0  0.30975143
6   12  2  57 200   0  1   0  45  19   0  0.40476190
8   12  0  38   2 251  0   0   0   1   0  0.17434211
10  12  0  10   3   0 10   0   0   0   0  0.71428571
11   0  0   1   0   0  0 192   4   1   0  0.03030303
12   1  0  37  19   0  1   2 293  35   0  0.24484536
14   0  0  24  14   1  0   0  30 619   0  0.10029070
17   0  0   0   0   0  0   0   0   0 350  0.00000000

plot(rf_model2) #modelo 2

Evaluación del modelo 1.

pred_rf_model1<- predict(rf_model1,newdata = eva[,-1])


cm <- confusionMatrix(data = pred_rf_model1, as.factor(eva$layer))
cm

Confusion Matrix and Statistics

          Reference
Prediction   2   3   5   6   8  10  11  12  14  17
        2  372  15  80   3   7   3   0   0   1   0
        3    2  13   1   3   0   0   0   0   0   0
        5   85   3 317  18  19   9   1  12   3   0
        6    3   1  24 102   1   1   0  10   7   0
        8    3   0   8   0 104   0   0   0   0   0
        10   0   0   0   0   0   2   0   0   0   0
        11   0   0   0   0   0   0  80   3   0   0
        12   0   0  13  11   0   0   3 123  13   0
        14   1   0   6   7   0   1   2  19 272   0
        17   0   0   0   0   0   0   0   0   0 150

Overall Statistics
                                          
               Accuracy : 0.7925          
                 95% CI : (0.7737, 0.8103)
    No Information Rate : 0.2406          
    P-Value [Acc > NIR] : < 2.2e-16       
                                          
                  Kappa : 0.7515          
                                          
 Mcnemar's Test P-Value : NA              

Statistics by Class:

                     Class: 2 Class: 3 Class: 5 Class: 6 Class: 8 Class: 10
Sensitivity            0.7983 0.406250   0.7060  0.70833  0.79389  0.125000
Specificity            0.9259 0.996850   0.8992  0.97379  0.99391  1.000000
Pos Pred Value         0.7734 0.684211   0.6788  0.68456  0.90435  1.000000
Neg Pred Value         0.9354 0.990094   0.9102  0.97651  0.98518  0.992765
Prevalence             0.2406 0.016520   0.2318  0.07434  0.06763  0.008260
Detection Rate         0.1920 0.006711   0.1637  0.05266  0.05369  0.001033
Detection Prevalence   0.2483 0.009809   0.2411  0.07692  0.05937  0.001033
Balanced Accuracy      0.8621 0.701550   0.8026  0.84106  0.89390  0.562500
                     Class: 11 Class: 12 Class: 14 Class: 17
Sensitivity            0.93023   0.73653    0.9189   1.00000
Specificity            0.99838   0.97740    0.9781   1.00000
Pos Pred Value         0.96386   0.75460    0.8831   1.00000
Neg Pred Value         0.99676   0.97520    0.9853   1.00000
Prevalence             0.04440   0.08622    0.1528   0.07744
Detection Rate         0.04130   0.06350    0.1404   0.07744
Detection Prevalence   0.04285   0.08415    0.1590   0.07744
Balanced Accuracy      0.96431   0.85696    0.9485   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_rf_model2<- predict(rf_model2,newdata = eva[,-1])


cm <- confusionMatrix(data = pred_rf_model2, as.factor(eva$layer))
cm

Confusion Matrix and Statistics

          Reference
Prediction   2   3   5   6   8  10  11  12  14  17
        2  373  17  76   4   8   2   0   0   1   0
        3    0  11   0   0   0   0   0   0   0   0
        5   87   4 320  22  19  10   1  14   2   0
        6    2   0  24 101   1   1   0   9   4   0
        8    3   0   8   0 103   0   0   0   0   0
        10   0   0   0   0   0   2   0   0   0   0
        11   0   0   0   0   0   0  80   2   0   0
        12   0   0  14  11   0   0   3 122  14   0
        14   1   0   7   6   0   1   2  20 275   0
        17   0   0   0   0   0   0   0   0   0 150

Overall Statistics
                                          
               Accuracy : 0.7935          
                 95% CI : (0.7748, 0.8113)
    No Information Rate : 0.2406          
    P-Value [Acc > NIR] : < 2.2e-16       
                                          
                  Kappa : 0.7523          
                                          
 Mcnemar's Test P-Value : NA              

Statistics by Class:

                     Class: 2 Class: 3 Class: 5 Class: 6 Class: 8 Class: 10
Sensitivity            0.8004 0.343750   0.7127  0.70139  0.78626  0.125000
Specificity            0.9266 1.000000   0.8931  0.97713  0.99391  1.000000
Pos Pred Value         0.7755 1.000000   0.6681  0.71127  0.90351  1.000000
Neg Pred Value         0.9361 0.989097   0.9115  0.97604  0.98464  0.992765
Prevalence             0.2406 0.016520   0.2318  0.07434  0.06763  0.008260
Detection Rate         0.1926 0.005679   0.1652  0.05214  0.05318  0.001033
Detection Prevalence   0.2483 0.005679   0.2473  0.07331  0.05885  0.001033
Balanced Accuracy      0.8635 0.671875   0.8029  0.83926  0.89008  0.562500
                     Class: 11 Class: 12 Class: 14 Class: 17
Sensitivity            0.93023   0.73054    0.9291   1.00000
Specificity            0.99892   0.97627    0.9775   1.00000
Pos Pred Value         0.97561   0.74390    0.8814   1.00000
Neg Pred Value         0.99677   0.97462    0.9871   1.00000
Prevalence             0.04440   0.08622    0.1528   0.07744
Detection Rate         0.04130   0.06298    0.1420   0.07744
Detection Prevalence   0.04233   0.08467    0.1611   0.07744
Balanced Accuracy      0.96458   0.85341    0.9533   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_rf_model1 <- raster::predict(roma_rst_lan, model=rf_model1)
plot(classi_rf_model1, main="Clasificación Random Forest Modelo 1",col = cols)

Predicción y visualización del modelo 2.

classi_rf_model2 <- raster::predict(roma_rst_lan, model=rf_model2)
plot(classi_rf_model2, main="Clasificación Random Forest Modelo 2",col = cols)