Development of Sweet Tamarind Sorting System by Image Processing Technique

Abstract

The sweet tamarind from Phetchabun province is classified as a Geographical Indication (GI) product and can earn billions of baht in annual revenue. It is imperative to categorise items based on their price, quality, and size. In the present day, the task of categorising sweet tamarinds needs individuals with expertise. When dealing with a substantial amount, it requires significant time and effort. The researcher proposes to create a sorting system for sweet tamarind using image processing techniques. The system will classify the sweet tamarind into categories such as sweet tamarind with flaws, normal tamarind size 1, normal tamarind size 2, and normal tamarind size 3. This will be achieved by employing YOLO version 8 with n, s, and m algorithms. The dataset consists of 3,000 photographs of sweet tamarind, specifically Srithong and Srichompoo varieties. These images were used to carry out a test with a ratio of 80:20, meaning that 80% of the images were used for training and 20% for testing. The dataset was separated into 5-Fold Cross Validation, which involves splitting the data into five subsets and doing the test several times, each time using a different subset as the test set. The results were then averaged. Following performance testing, the researcher opted for the n model, which has a maximum speed of 7.28 frame/s, a mAP value of 0.84, a precision value of 1.0, a recall value of 8.9, an F1-Score of 0.75, and the smallest size on disc, 7.52 Mb. This model was then utilised to create a web application for predicting the quality of sweet tamarinds. This technology helps farmers in efficiently and precisely categorising sweet tamarind, while also providing customers with tools to enhance their trust and confidence in the product.