Botanic Alchemy: Cultivating Insight through Extensive Plant Disease Dataset and Reweighted Visual Analysis

Diagnosing plant diseases research tackles the vital challenge of identifying plant diseases in agriculture by focusing on the intricate complexities found in images displaying diverse symptoms, scattered lesions, and complex backgrounds. To aid this endeavor, a substantial dataset of 220,592 images is assembled. The proposed method uses a reweighting strategy that targets both visual regions and loss. Initially, weights for segmented patches are calculated based on their discriminative potential obtained from cluster distributions. These weights are then seamlessly integrated into the training process, assigning them to individual patch-label pairs during weakly-supervised training, aiding in recognizing disease-specific regions. Features extracted from this trained network create a comprehensive feature representation. This integration incorporates the GRU algorithm to encode the sequence of weighted patch features. Thorough evaluations on the newly curated dataset and a public dataset demonstrate the effectiveness and superiority of this methodology. This research significantly advances the field of plant disease recognition in image processing, opening doors for further innovative exploration.