Preferred Language
Articles
/
alkej-1250
Advanced Hybrid Convolutional Neural Network for Leaf-Based Plant Disease Detection
...Show More Authors

Accurate detection and classification of plant diseases are central to sustainable food production and the reduction in crop losses. Conventional identification protocols are often dependent on experts. They are time-consuming and difficult to scale. This paper proposes a novel Advanced Hybrid Convolutional Neural Network (AHCNN) model combining the attention mechanism and simplified convolution mechanism. This model can obtain high accuracy and low computational complexity. It is trained using a high-resolution leaf dataset that includes 1532 leaf images, separated into 1322 training, 150 validation and 60 test samples, with the goal of a classification of specimens into three categories: healthy, powdery and rusty. Its architecture uses squeeze-and-excitation (SE) blocks and a spatial attention mechanism, which, together, can provide enhanced feature extraction and improve the interpretability of models. Results showed that the model achieved a validation accuracy of 98% and a test accuracy of 98.33%. With a parameter number of only 0.4 million, the proposed architecture provides a lightweight solution that performs much better than conventional deep learning frameworks in terms of computational efficiency. These attributes make AHCNN an interesting candidate for real time, drone-based detection of plant diseases as part of precision agriculture systems.

View Publication Preview PDF
Quick Preview PDF