Disease Identification in Tomatoes Using AI (DITA)

India’s agricultural sector, employing 42.3% of the population and contributing 18.2% to national GDP in 2023-24, produces abundant food and supports national food security. Post-harvest vegetables and fruits, such as tomatoes, have limited shelf life, with ripe tomatoes lasting just 10–12 days. Effective storage at optimal temperatures is crucial, and farmers must monitor produce health and sell before spoilage, but manual sorting is labor-intensive. DITA provides a solution to monitor the health of specifically tomatoes. It uses bounding box results from YOLOv12(small) and MobileNet v3 object detection models to predict labels for each frame. The Simple Online and Real-time Tracking (SORT) algorithm is used to combine and process the detections from Yolov12 and Mobile Net model, sorting them by confidence score in descending order. Intersection over Union (IoU) is used to determine overlapping bounding boxes, and boxes with low confidence scores are eliminated. Remaining boxes are classified using thresholds of 0.5 and 0.6 for YOLOv12 and 0.5 for MobileNet v3. YOLOv12 accurately classifies RIPE, UnRipe, and Rotten classes, though ‘Damaged’ accuracy is 50%. Thresholds of 0.5 and 0.6 are applied to YOLOv12 confidence scores when combining results with MobileNetV3. For object tracking with SORT, IoU thresholds of 0.3 and 0.5 associate detections across frames. These thresholds are also applied when merging bounding boxes and maintaining object identity during tracking. Since MobileNetV3 operates as a black-box model, Grad-CAM + + is employed to provide visual explanations of its predictions.