DETEKSI KESEGARAN DAGING SAPI MENGGUNAKAN YOLO

Abstract

The main goal of this research is to know whether the model YOLO could be used to detect the freshness of beef, in which the problem is also a classification problem. The version of YOLO that is used in this research are YOLOv8n, YOLOv7-tiny, and YOLOv5n. Three testing scenario is used in this research. In first testing scenario, the models are tested using dataset of new images that were never seen in the training process. In the second scenario, the models are tested using dataset of new images that have gone through pre-processing, namely cropping and resizing. In the third scenario, the models are tested using new images that have gone through HSL (Hue, Saturation, and Luminance) adjustment, in which the result of the detection will be compared with the result of the detection of the very same image without HSL adjustment. The accuracy metric is used to measure the performance of pure classification task of the three models, and the metric Mean Average Precision is used to measure the performance of object-detecting task of the three models. From the result of the research, the cropping and the resizing process are very influential to the end result of the performance of the three models, in which the performance of the three models in the second testing scenario is improved by more than 700% compared to the first testing scenario. The process of HSL adjustment is also found to be very influential to the result of the detection by the three models. The model with the best performance in terms of pure classification task is the YOLOv8n, with 5.66% accuracy in the first testing scenario and 47.32% in the second testing scenario. The model with the best object-detecting performance in the first testing scenario is the YOLOv8n with an mAP 0.5 of 0.044, while the model with the best performance in the second testing scenario is the YOLOv7-tiny with an mAP 0.5 of 0.457. From the result, it can be concluded that the three YOLO version is not yet capable of becoming the solution for the problem of detecting beef freshness.