Applying image segmentation method to estimate the number of irregular overlapping objects – Taking the use of YOLOv8 to estimate the quantity of pineapple angiosperms in open farmland as an example

Chien-Hung Lai, Yi Lin, Yuh-Shyan Hwang, Po-Chun, Lin

Abstract:

This study proposes a solution that combines drones and the YOLOv8 algorithm to solve the problem of difficult object detection owing to irregular and overlapping leaves of pineapple angiosperms. The leaves of pineapple angiosperms are arranged in spirals, lamellas and naturally spread out in all directions to increasing the area for photosynthesis. But object detection technology faces the challenge of overlapping distinctions during labeling for training. To solve this problem, this study uses drones to capture high-resolution field images and uses the YOLOv8 algorithm to perform image segmentation and object detection. It combines data argumentation and segmentation technologies to effectively improve the model's ability to detect overlapping objects.
Experimental result show that the solution proposed in this study can quickly and precisely estimate the quantity of pineapple
angiosperms, the detection accuracy can reach more than 95% in overlapping situations. The solution provides efficient and reliable
technical framework for automated data management in precision agriculture. In addition to overcoming the time-consuming and
labor-intensive shortcomings of traditional manual estimation methods, the solution also significantly improves the efficiency of
agricultural data collection and processing.