Deep feature based rice leaf disease identification using support vector machine

•This study evaluates the performance of 13 CNN models for rice disease identification in transfer learning and deep features plus SVM approach.•The statistical analysis results, deep features of resnet50 and SVM classification model is superior compared to other models.•A comparative analysis is carried out among all classification models based on CNN and traditional methods.•The performance of mobilenetv2 plus SVM is comparable enough with the resnet50 plus SVM.•In addition, the study is carried four varieties of rice leaf diseases with a dataset of 5932 on-field images. Features are the vital factor for image classification in the field of machine learning. The advancement of deep convolutional neural network (CNN) shows the way for identification of rice diseases using deep features with the expectation of high returns. This paper introduced 5932 on-field images of four types of rice leaf diseases, namely bacterial blight, blast, brown spot and tungro. In addition, the performance evaluation of 11 CNN models in transfer learning approach and deep feature plus support vector machine (SVM) was carried out. The simulation results show the deep feature plus SVM perform better classification compared to transfer learning counterpart. Also, the performance of small CNN models such as mobilenetv2 and shufflenet was examined. The performance evaluation was carried out in terms of accuracy, sensitivity, specificity, false positive rate (FPR), F1 Score and training time. Again, the statistical analysis was performed to choose the better classification model. The deep feature of ResNet50 plus SVM performs better with F1 score of 0.9838. The fc6 layer of vgg16, vgg19 and AlexNet have better contribution towards classification compared to fc7 and fc8. Further, the F1 score of CNN classification models was compared with other traditional image classification models such as bag-of-feature, local binary patterns (LBP) plus SVM, histogram of oriented gradients (HOG) plus SVM and Gray Level Co-occurrence Matrix (GLCM) plus SVM.