Sensitive crop leaf disease prediction based on computer vision techniques with handcrafted features

Agricultural production is considered the primary source of the economy of many countries. Tomato and Potatoes are the most sensitive and consumable vegetables worldwide. However, during the growth of these crops, they suffer from many leaf diseases, which lead to loss of productivity and economy of the farmers. Many farmers detect and find plant diseases that are more time-consuming, expensive, and require expert decisions following the naked eye method. Therefore, early and accurate diagnosis of Tomato and Potato crops leaf diseases plays a vital role in sustainable agriculture. So, this research paper proposes an efficient leaf disease classification model based on computer vision techniques. The proposed Adaptive Deep Neural Network (ADNN) leaf disease classification method is a hybrid model which combines an optimized long short-term memory (OLSTM) and convolution neural network (CNN). The weight values supplied in the LSTM classifier are optimally selected using the Adaptive Raindrop Optimization algorithm. The handcrafted features are extracted from the segmented image and fused with the hybrid deep neural network to improve the classifier performance. The ADNN method consists of five steps: preprocessing, feature extraction, segmentation, handcrafted feature extraction, and classification. At first, the images are given to the preprocessing stage to remove the noise from leaf images. Then, the image-affected portion is segmented using an enhanced radial basis function neural network. After the segmentation process, the segmented image is given as an input to the adaptive deep neural network (ADNN) that classifies various types of diseases in the Potato and Tomato leaves. The efficiency of the ADNN model based on the OLSTM-CNN approach is determined concerning multiple metrics, namely Accuracy, Precision, Recall, F-measure, Specificity, and Sensitivity. The ADNN model achieved the best Accuracy of 98.02% for Tomatoes and 98% for Potatoes. The ADNN is compared with existing state-of-the-art CNN, LSTM, ResNet50, and MobileNet techniques. The performance analysis proved that the ADNN model improved efficiency in terms of all metrics and methods.