Real-Time Banana Ripeness Detection and Classification using YOLOv8

Due to farmers in the Philippines having to utilize manual methods to determine the grade and quality of freshly harvested bananas, post-harvest classification and segregation accounts for between 3 % and 30% of food waste that is linked to bananas. This study focused on real-time banana object recognition and classification via a webcam using neural networks. The principles of computer vision and object recognition and how they relate to fast algorithms like CNN and YOLO are a major emphasis of this study. The researchers used the YOLOv8 algorithm to develop their model since the appropriate selection of the YOLO model determines both the accuracy and processing speed of the photographs. The method is intended to produce a detected and classified level of banana ripeness and be displayed through a live feed from the webcam. Using the confusion matrix, the model was able to detect and categorize objects with an accuracy of about 90% across all classes on the testing dataset of static images. For real-time detection of the model, samples were placed against a blue background given that the yellow and green colors of banana samples are a complementary pair according to the sRGB color wheel. Banana samples were tested both in bunches and as individual samples. However, the model was only 80% accurate during real-time individual sample testing.