A novel framework for soybean leaves disease detection using DIM-U-net and LSTM

Soybean leaf disease is one of the major problems that reduces the agricultural productivity. Detection of the soybean leaf diseases based on their category is a strategy that can annihilate them and result in an increased productivity. Moreover, erroneous detection of disease can lead to inappropriate treatments that can affect the healthy leaves. To overcome this problem, proper detection and classification techniques have to be implemented. To accurately detect and classify the diseases of the soybean leaf, Dense Inception Module based U Net Segmentation- (DIM-U-Net) with a Sparse Regularized Auto Encoder (SR-AE) and Long Short-Term Memory (LSTM) for classification is proposed. In this study, the soybean leaf diseases from the images are detected based on the DIM-U-Net which is a deep learning model that segments the image with encoding and decoding processes whose output which is the segmented image is sent to the SR-AE for feature extraction. Finally, an LSTM classification is applied to classify the leaves into three classes namely, Angular Spot, Bean Rust and Healthy. This avoids the wrong detection of the category of the leaf disease. The DIM-U-Net detected the diseased leaves and the various performance metrics such as Classification Accuracy Rate-CAR, sensitivity, specificity, precision, F1 score and AUC values have been identified in the classification of the leaves and by comparison the proposed DIM-U-Net model outperforms the existing methods.