Cotton leaf net: DeepLabv3+ powered semantic segmentation framework for precise cotton plant leaf disease detection in smart agriculture

Accurate and effective plant leaf segmentation is essential for crop health monitoring and early disease detection in precision agriculture. In this study, a promising methodology to segment cotton leaves is presented using DeepLabv3+ architecture, a well-known semantic segmentation model with high accuracy and edge-preserving abilities. The proposed methodology employs advanced pre-processing techniques as part of the system to enhance the image quality prior to segmentation, utilizing methods such as vegetation indexing, colour space transformation, and noise reduction. The model employs DeepLabv3+, with a ResNet-50 backbone and aurous spatial pyramid pooling (ASPP) to deal with multi-scale leaf structures, even in noisy field conditions. Completeness is verified by evaluating accuracy and pixel-wise F1 score in rural areas of India, where cotton is the most widely grown cash crop and forms much of the textile and agricultural industries. Cotton plant health and productivity can be greatly affected by a multitude of leaf diseases and pest infestations, such as cotton leaf curl virus (CLCuV) and whiteflies. Consequently, minimizing crop loss and ensuring sustainable farming practices are contingent on the prior identification and accurate assessment of crop conditions. The Dice coefficient emphasizes the effectiveness of the use of the proposed method that surpassed conventional models such as U-Net and SegNet, achieving an overall accuracy of 96.5% and a Dice score of 0.92. The method not only guarantees accurate location of boundaries but also performs well with complex texture and high contrast of cotton leaves, making it especially suitable for identification of chick’s disorders such as leaf curl and white fly infestation. The technology not only enables rapid and accurate base interventions for crops, it makes a critical contribution to smart farming applications related to monitoring at both crop greenhouse and field spatial scale.