Tomato Fruit Defect Detection System Using SUSAN Edge Detection, Statistical Feature Extraction, and CNN

Abstract

This research aims to address the problem of defect detection in tomatoes, which often compromises product quality in the agricultural industry. The difficulty in detecting defects automatically and accurately is a major challenge, so an efficient and effective method is needed. For this reason, a detection system was created by combining SUSAN edge detection method, statistical feature extraction, and Convolutional Neural Network (CNN). The SUSAN method was chosen for its reliability in detecting edges well, which is important for identifying defective areas in tomatoes. The process starts with edge detection using the SUSAN method, followed by statistical feature extraction such as mean value, standard deviation, minimum value, and maximum value of pixel intensity in tomato images. This data is then used to train the CNN model, which achieves a training accuracy of 97.50% and a test accuracy of 90%. From testing 50 tomato samples, CNN accuracy of 96%, precision of 96%, and recall of 100% were obtained. These results show that this system works well in detecting defects in tomatoes. Thus, this system is expected to improve the quality of tomato products and support the quality standards of the agricultural industry.