Enhanced Jute Pest Identification Using EfficientNetB1 with Median Attention Mechanism

Jute is a significant economic crop, and accurate identification of jute pests can assist farmers in implementing effective control measures and mitigating yield losses. However, traditional manual identification methods are inefficient, highlighting the need for rapid and precise pest classification tools. This paper presents a jute pest image classification model based on EfficientNetB1, incorporating a median-enhanced spatial and channel attention mechanism. The proposed model leverages a convolutional neural network architecture capable of automatic feature extraction, facilitating both classification and prediction of images. The experimental dataset consisted of 6212 images representing 17 distinct jute pest species, which were divided into training, validation, and test sets at a ratio of 3:1:1 by random. To enhance feature extraction capabilities, we introduced a median-enhanced spatial and channel attention block into the EfficientNetB1 framework. This module, termed MESC-EfficientNetB1, integrates the benefits of both channel and spatial attention mechanisms, allowing for the capture and fusion of multi-scale features. Experimental results demonstrate that MESC-EfficientNetB1 showed excellent results compared to some other models. These findings underscore the high accuracy of the enhanced module, leveraging the lightweight nature of the model and advanced recognition technology to accurately classify jute pests. This study offers a valuable reference for the development of methodologies for identifying jute pests.