Incorporating Dilation Convolution into Mask Region Convolution Neural Network for Advanced Fruit Classification and Freshness Evaluation

Fruit supply and storage chains play a significant role in the food economies of all countries that export and consume these crops. Reducing the losses caused by the damage to these crops and organizing their marketing, relying on artificial intelligence methods, has begun to receive increasing attention from researchers. To ensure food safety and meet consumer expectations. This paper investigates the integration of dilation convolution into the Mask R-CNN framework to improve the accuracy of produce classification and detection of freshness. This paper seeks to address the challenges associated with traditional methods of fruit quality assessment, such as visual inspection and manual grading, which are time-consuming and subjective. To ensure the robustness of our approach, we collected a diverse set of complex images under various lighting conditions, and we implemented preprocessing techniques, such as histogram equalization and color correction, to enhance image quality. Using instance segmentation allowed us to distinguish and classify overlapped objects within each image. The practical results demonstrate that integrating dilation convolution improves the fruit classification, with an accuracy of 97.8% over the baseline Mask R-CNN. Real-time images and video were applied using the trained model, proving its practical utility in dynamic environments. It demonstrates the potential for direct object detection and classification by proposing a valuable tool to enhance productivity and ensure product quality.