Early Detection of Leaf Venation Damage Using Smart Deep Learning with C-LSTM and GradCAM

The study of inherited characteristics and venation pattern in leaves has recently been studied significantly because of its importance in the study of plant health and early detection of damage. Reticulate or parallel leaf venation, regardless of more recent loss, is an important indicator of how ‘healthy’ a plant is and how resilient it is. An innovative method for detecting damage to leaf venation is proposed using extensive segmentation and detection methods to rapidly detect damage within leaf venation. The FH algorithm, a non-parametric, reliable and fast segmentation method, is applied to the suspicious area of the leaf. The GradCAM (Gradient-weighted Class Activation Mapping) method is then used to locate the RoI as a visualization tool to guide the detection process. A convolutional-LSTM (C-LSTM) model is then used to detect leaf venation damage that utilizes convolutional and LSTM techniques to effectively analyze spatial and temporal data. The empirical results of this study reveal that the proposed model performs significantly better than the traditional model, and it was able to achieve more than 98.64% detection accuracy. The approach demonstrates significantly improved detection versus other existing models in detecting early signs of leaf venation damage. The proposed segmentation and detection in leaf venation for plant protection focuses on the vital importance of early damage detection in plant protection. This approach provides a basis for applications in plant health monitoring beyond framing the problem, which could aid in improving agricultural practices and the speed of intervention to maintain plant health.