Comparison and Analysis of YOLOv5 and YOLOv4-Tiny in Fruit Sample Detection

The Raspberry Pi is commonly used in embedded object detection applications due to its support with frameworks frequently used in object detection algorithms and being a low-cost platform with sufficient power to run these algorithms. Among numerous object detection algorithms implemented on the Raspberry Pi, some examples include RCNN, MobileNetv2, and the YOLO algorithm. This article will focus specifically on the YOLO algorithm. In this article, two implementations of the YOLO algorithm: YOLOv4-Tiny and YOLOv5, will be compared regarding their accuracy and speed in fruit sample detection. From testing, it was found that both algorithms can accurately detect the fruit samples in a given image, YOLOv5 performed noticeably better in object localization through bounding boxes and speed. Specifically, YOLOv5 detected the fruit sample in the given image in 11.7559 seconds on average, making it 3.87 times faster than YOLOv4-Tiny, according to the results, while having a higher average confidence value.