Tomato stem scar detection using C3Ghost-based improved YOLOv5 model

The tomato is an important vegetable worldwide due to its increasing demand, which includes household cooking, industries, and restaurants.A specific structure is followed for arranging tomatoes in food industries, where the tomatoes are usually placed in packets with the stem scar on top.A stem scar is an area on the top of a tomato fruit whose shape, smoothness, and diameter determine tomato freshness.As a result, it is considered a reliable indicator of tomato quality.So, identifying the stem scar from different angles is a challenging task due to texture disturbance, partial occlusion, etc. Different studies mentioned that the above challenges can be overcome by applying deep learning approaches.Therefore, this research proposes a modified YOLOv5 algorithm to identify the stem scar of tomatoes by adapting a C3Ghost module in the backbone part to reduce computational costs.To enhance the overall detection rate, two activation layers of leaky ReLU with a negative slope of 0.1, a non-maximum suppression, and an extra convolution module with padding are added to the neck region.A self-generated dataset is prepared and used to test and compare the proposed modified YOLOv5 method with the existing YOLOv5 algorithm.In the case of performance evaluation, the mean average precision (mAP) of the model is 98% which is 4.6% better than the existing model.Hence, it can be used for industrial tomato grading at a low cost.