• Development of a Tomato Fruit Anomalies Detector for a Small Greenhouse Drone Application

Abstract

Tomatoes are an important commercial crop. Tomatoes are popular in many parts of the world because they are easy to grow and have a high nutritional value. The fact that tomato fruits are prone to disease, even when grown in greenhouses, is a major challenge in large-scale tomato production. Agricultural drones are now being used to construct precision farming systems that can automatically detect irregularities in tomatoes. The main challenge with using a drone in a greenhouse is that the drone must be tiny enough to carry a restricted load, including an onboard computer. Because of this, the algorithm used to detect tomato fruit anomalies must be compact and efficient. The tomato fruit abnormalities detecting method was proposed in this research. The method involved with CNNs and image processing algorithms. First the location of each tomato in tomato vines were identify and localized by using YOLOv4-tiny CNNs model, then, each tomato fruit was cropped. The background of the tomato fruit was removed by graph-like structure called GrabCut. Finally, all major defects were detected separately by the segmentation-based color thresholding technique. The experiments were conducted. The result showed that YOLOv4-tiny could detect tomato fruits with 98.68% AP and the color thresholding successfully detected the tomato anomalies.