Improving Orange and Lemon Object Detection Using YOLOv5 and Image Augmentation

One of the crucial techniques that can be used to augment the dataset at the pre-processing stage is augmentation techniques. Widely used augmentation techniques are the transformation of the original image, change of scale variation, rotation, setting the brightness level of the image, and many more. Image augmentation aims to increase the robustness and capability of the built model in generalizing object detection using the YOLOv5 algorithm. In this study, we evaluate several image augmentation techniques with a combination of epoch configurations in the YOLOv5 algorithm. Our results show that the image inversion technique can consistently improve the performance of YOLOv5 compared to raw images. On the other hand, based on the analysis of execution time with larger epoch configurations, more time is needed for model building. Thus, the trade-off between the time needed to execute and the accuracy of the model's performance in detecting objects allows decision-making according to the needs.