A Composite Meta Model for the Identification of Cotton Pathologies Utilizing an IoT-Enabled Framework and Stacked Generalization Learning Methodology

This research proposes a novel framework for predicting cotton plant diseases using IoT, deep learning, and meta-heuristic optimization techniques. High-definition images of cotton leaves are captured in the field, processed through IoT, and enhanced using a Probabilistic Hybrid Wiener Filter. The Modified Dilated U-Net segments pathological regions, while features are extracted using Improved Local Binary Pattern (LBP), Gray Level Co-Occurrence Matrix (GLCM), and Gray Level Run Length Matrix (GLRLM). Feature dimensionality is reduced by the Binary Guided Whale-Dipper Throated Optimizer. The classification uses an ensemble of deep learning models—EfficientNet-B7, ResNet50, VGG19, DenseNet121, and InceptionV3—optimized by Harris whale optimization to determine weight coefficients. The system accurately detects diseases like Army Worms, Powdery Mildew, and Bacterial Blight with 99.66% accuracy. This IoT-enabled framework provides efficient real-time disease detection, benefiting cotton farmers and the textile industry. A field study was conducted in the summer (Kharif) season of 2022–23 in North Maharashtra region to assess cotton cultivation utilizing IoT sensor data analyzed within the ThingSpeak IoT framework. The proposed methodology, leveraging a dataset of the images of cotton leaves demonstrate a remarkable precision rate of 99.66%. The amalgamation of IoT sensor data with deep learning methodologies enables the early prompt identification of diseases in cotton plant leaves. The suggested ensemble framework demonstrates enhanced efficacy in comparison to alternative models.