Designing and improving the automatic detection system for fruit defects using Attention-Enhanced CNN

Abstract

This study presents an innovative approach to improving the accuracy of fruit quality assessment. We propose the integration of an enhanced attention mechanism with Swish activation and a residual connection within a Convolutional Neural Network (CNN) architecture. Our method addresses the crucial task of accurately categorizing the quality of six distinct types of fruits. By leveraging attention mechanisms to highlight relevant features and utilizing a residual connection to facilitate hierarchical learning, our model achieves an impressive accuracy rate of 96.46% in fruit quality classification. The combination of attention-driven processing and the residual connection contributes to a more nuanced and discriminative understanding of fruit quality attributes. This research holds significant promise for revolutionizing fruit quality assessment practices across various industries, improving efficiency, and ensuring that only the highest-quality products reach consumers.