Asynchronous Overlapping: An Image Segmentation Method for Key Feature Regions of Plant Phenotyping

Abstract

Highlights Asynchronous overlapping—an automatic image acquisition method for key feature regions of plant phenotypes. The distance from the plant to the camera can be characterized by the brightness in the grayscale image. Asynchronously acquire daytime RGB and nighttime grayscale images of the plant to use the proposed algorithm. In the test of the plant images, the IoU is 0.8497, reaching a similar level of interactive algorithms. Abstract. Acquiring and describing plant phenotyping is an important proposition in botany and agronomy research. In this study, a computer vision-based asynchronous overlapping segmentation algorithm is proposed for automatic image acquisition of key feature regions of plant phenotyping. Firstly, day-time RGB and night-time grayscale images of infrared light filling the crop body at the same angle are asynchronously obtained using a common closed-circuit television surveillance camera. Then, thresholding and morphological filtering of grayscale images are conducted to extract the initial region contours. With this as a precondition, the algorithm adaptively finds edge paths of key feature regions in daytime RGB images. In the test of the cherry plant image, the intersection over union (IoU) of the algorithm to segment the key feature regions is 0.8497, reaching a similar level of interactive algorithms that require human involvement. The proposed method has low cost, high segmentation accuracy, and strong applicability. The proposed method can independently realize the acquisition of the key feature regions of plant image phenotypes and can be applied to large-scale agricultural production.