THE JOURNAL OF KMUTNB Volume 33, No. 02, Month APRIL, Year 2023, Pages -
Digital weighing scale with fruit and vegetable identification using deep learning technique
Pisit Wisutmetheekorn, Noppawut Sudsawat, Nakarin Ruksawong
Abstract Download PDFWeighing is a process before packing fruits and vegetables for sale in supermarkets. A staff will put the product to be weighed on the digital scale and enter the product code. After that the machine will display product name, price, and weight. However, error might occur from entering the wrong product code. This research aims to develop digital weighing scale that an image data is imported from the camera to identify the types of fruits and vegetables instead of entering the product code. The system consists of Raspberry Pi computer board, Pi camera V2.1 for image acquisition, load cell for weighing, HX-711 analog to digital converter module and a touch screen. Software is developed by Python version 3 and the YOLOv3-tiny algorithm is used through the Darknet library to build a model for image recognition. For testing the system, five types of fruit and vegetable were used to train the model. The pictures in training process included 379 pictures of banana, 277 pictures of carrot, 232 pictures of grape and 443 pictures of onion. The model was built by 15000 epochs of training with 0.1623 of an average loss value and the values of precision, recall and F1-score are 1.00, 0.99 and 0.99 respectively. From the experimental results, the system can identify banana, carrot, grape and onion with 100% accuracy when fruits or vegetables are placed without overlapping. However, error may occur if fruit or vegetable are overlapping. The weighing error of the developed weighing scale is less than 10 grams.
Deep Learning; Image Recognition; YOLO Algorithm