PLANT DISEASE DETECTION USING A HYBRID DEEP CNN MODEL WITH ATTENTION MECHANISM

Abstract

Early and accurate recognition of plant leaf diseases is crucial in order to protect crop yields and take preventive measures in time.  Guess what? However disease identification remains a challenge due to high similarity between classes’ complex morphological variation and limitations in training data quality. This paper presents a unified hybrid deep learning framework that integrates convolutional neural networks residual and primal , primal units convolutional block monitoring mechanisms (CBAM) supports transformer-based global feature extraction and support vector machine (SVM) classification to optimize and improve disease detection in multiple plant species. Generative models such as loop-adapted GANs are used to deal with data imbalance and improve generalization.  Seriously further innovations including frequency-domain monitoring networks (FdaNet) and hybrid, hybrid monitoring networks (HaNet) are improving feature extraction in complex crop environments with small-scale lesion patterns. Experiments on different datasets covering corn potato tomato and citrus showed high performance with recognition accuracy reaching 99.55% for multiple plant disease classification 99.45% for tomato leaf diseases and 98.83% for citrus, citrus disease detection. The proposed models , models have fewer parameters shorter training time and are robust to real-world scenarios and can be successfully applied in web-based and mobile applications for real-time disease and illness diagnosis. These results show an efficient and scalable solution for the early detection of plant diseases in agricultural fields.