Progressive Expression of Bean Leaf Lesions: A Comprehensive Analysis Using Spatio-Temporal Disease Classification Solutions Based on CNN and LSTM Networks
Deepak Upadhyay, Manika Manwal, Ajay Gupta, Vinay Kukreja, Rishabh Sharma
Abstract
The appearance of Bean leaf lesions can lead to major damage and losses in crop production, requiring innovative approaches for early detection. This study proposes a new hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model for bean leaf lesion detection, especially in the classification of severity levels. It is based on a highly maintained database with 7000 images representing lesions at different stages. The proposed model presents a high overall accuracy of 98.52% which highlights its ability to distinguish lesions on each of the four severity levels – mild, moderate, severe, and critical. The architecture of the model comprises CNN to extract features from leaf images based on spatial awareness and LSTM for temporal lesion detection that captures how the pattern changes over time. The evaluation of performance parameters such as precision, recall, and F1-score helps augment the analysis conducted on accuracy since they provide insights into what works well within the model for further development. The addition of a confusion matrix depicts the forecast performance of the model and more specifically what severity classes present where the accuracy is good or needs attention. Not only does this study add to the state-of-the-art in disease detection of crops but also contributes towards a wider discussion on utilizing artificial intelligence applied for precision agriculture, reasonable decision making, and reliable crop management.