Automated Pest Classification and Pesticide Suggestion using Deep Learning Algorithm

Abstract

The correct and real-time identification of crop pests in agricultural environments is vital for precision farming and effective pest management. This paper introduces an Attention-Based Compound Convolutional Neural Network (ABCCNN) model to improve the recognition of pests in complex field conditions. The proposed model integrates a Parallel Attention Mechanism (PAM) module with Residual Blocks (RBs) to improve feature extraction, localization, and classification. The PAM module focuses on the relevant pest features while suppressing the background noise so that the system can recognize the objects reliably in cluttered environments. The residual blocks facilitate better gradient propagation and convergence, thus making the network deeper and efficient. The proposed ABCCNN framework achieves accuracy and real-time performance better than state-of-the-art deep models with an optimum trade-off in computational efficiency as well as accuracy. Extensive experiments on pest datasets from reality demonstrate that classification accuracy, inference speed, as well as environmental variation robustness, such as lighting, occlusion, or complex backgrounds for conventional CNNs, are always superior to other existing models. Further, with the proposed model, the generalization of CNN-based pest recognition frameworks boosts up and enhances the possibility of large-scale deployment in smart agricultural systems. ABCCNN also contributes to early pest intervention strategies by providing fast, accurate pest detection, avoiding higher crop losses, and achieving sustainable agricultural practices. The model is planned to be further integrated into the edge AI devices in real-time on-field pest detection and decision support systems.