IoT-Enhanced Meta-Heuristic Hybrid Deep Learning Model for Predicting Cotton Leaf Diseases

Abstract

In the textile industry, cotton serves as a crucial raw material; however, diseases affecting cotton leaves can result in substantial financial losses for farmers. Conventional illness detection techniques are frequently costly, time-consuming, and inaccurate. Existing deep learning models can detect and classify affected leaves, but they face several limitations, including high error rates, excessive time consumption, a tendency for overfitting, and suboptimal performance. To overcome these issues, this work proposes a hybrid deep learning model with meta-heuristic support integrated with Internet of Things applications to efficiently classify cotton plant diseases. This creative concept seeks to give the textile sector and farmers a more precise and efficient solution. The proposed approach consists of two phases: first, high-resolution images of cotton leaves are captured using a Canon EOS 450D digital camera, and potential diseases are identified through IoT sensors. In the second step, advanced techniques like pre-processing, segmentation, feature extraction, feature selection, and classification are implemented. Disease segmentation is accomplished via the modified dilated u-net (MDU-Net) model. Feature selection utilising the Binary Guided Whale-Dipper Throated Optimizer (BGW-DTO) helps to identify the most relevant properties. Using the Harris Whale Optimization Method, the best weight coefficients for every classifier are found; next, a stacking ensemble model using the most recent deep learning approaches performs classification. In a collection of photos of cotton plant leaves, the optimal ensemble model shows a 99.66% classification rate, thereby precisely diagnosing a range of illnesses comprising Army Worms, Powdery Mildew, Bacterial Blight, Aphids, and Target Spots.