Abstract

In the realm of Indian agriculture, cotton stands out as a crucial commercial product. However, challenges persist in the form of various issues affecting the leaves, primarily stemming from elusive illnesses and pests not easily discernible by the naked eye. This paper leverages a deep learning methodology, specifically the Convolutional Neural Network (CNN), to devise a model aimed at enhancing the identification of cotton leaf diseases and pests. Notably, common afflictions such as bacterial blight, spider mite, and leaf miner are targeted in this study. The Kfold cross-validation technique is employed to partition datasets, contributing to the refinement of the CNN model's generalization. The study employs approximately 2400 specimens (600 photos per class) for training purposes, implemented using Python 3.7.3. The model is constructed using the Keras deep learning package, supported by TensorFlow, and developed in the Jupyter environment. Achieving a commendable 96.4 percent accuracy in recognizing classes of leaf diseases and pests in cotton plants, our model demonstrates practical utility for real-time applications. This success underscores the potential for integrating IT-based solutions to complement or augment traditional and manual methods in the ongoing quest for effective disease and pest detection in the agricultural domain.