Detection of external defects of tomato crop using appearance parameters by convolutional neural networks

Tomatoes, belonging to the Solanaceae family, are a vital vegetable cultivated worldwide, both in open fields and greenhouses. While field tomatoes are typically utilized for industrial purposes, greenhouse varieties are predominantly consumed fresh. Quality is a paramount factor influencing tomato consumption, hence this research endeavors to leverage novel technology to enhance quality assessment. Machine vision and deep learning systems were employed in this study to assess and categorize tomato quality. Samples were sourced in bulk from Tare-Bar Central Square in Tehran's 3rd District. Using a machine vision system illuminated by a ring light, samples were captured with a mobile camera, and the digital input data was processed using YOLOv7, a convolutional deep learning network. Training the YOLOv7 model necessitated 11 minutes and 60 epochs, culminating in an error rating of 0.017 with satisfactory outcomes. The average detection time per image, encompassing healthy, with calyx, semi-ripe, and defective tomatoes, was 0.048 seconds. Furthermore, the model achieved an impressive accuracy of 99.2 % and a recovery rate of 99.4 %. The findings of this project underscore the efficacy of the proposed model in automating tomato grading processes.