Authors

Abstract

The utilization of machine vision and image processing techniques enables the recognition of corn stalks in complex field conditions, thereby enhancing the efficiency and accuracy of corn stalk identification. Initially, images of corn stalks are captured using a camera, followed by Gaussian filtering to effectively reduce noise within the images. Subsequently, RGB and HSV color segmentation is applied to the corn stalk images to isolate the component V, which most distinctly differentiates the corn stalks from the background. Threshold segmentation is then conducted on component V, establishing a threshold range to identify and extract the corn stalks from the background. Morphological operations are employed to enhance the fullness of the image. Finally, grid division is performed, using the coverage area of the straw as a feature to assign density values to corn straw in various sub-regions, ultimately generating a distribution grade map of corn straw. The results indicate that when the straw coverage rate ranges from 25% to 80%, the discrepancy between the corn straw recognition rate obtained through threshold segmentation and the manually measured straw coverage rate is less than 5%, satisfying the requirements for subsequent real-time variable spraying of corn straw decomposition agents.

Abstract in Chinese

运用机器视觉和图像处理的方法可实现玉米秸秆田间复杂条件下的识别，提高玉米秸秆田间识别效率和精度。首先通过摄像头对田间玉米秸秆进行采集，将采集到的田间玉米秸秆图像进行高斯滤波处理，有效消除图像中的噪音；然后对玉米秸秆图像进行RGB和HSV颜色分割，找出玉米秸秆与背景区分最明显的分量V；再对分量V进行阈值分割处理，设定阈值范围，将玉米秸秆从背景中识别提取出来，运用形态学运算，使图像更加饱满；最后进行网格划分，以秸秆覆盖率面积作为特征，将不同子区域内的玉米秸秆密度进行赋值，生成玉米秸秆的分布等级图。结果表明：秸秆覆盖率为25%~80%时，通过阈值分割的玉米秸秆识别率与人工测量的秸秆覆盖率结果误差小于5%，能够满足后续玉米秸秆腐解剂实时变量喷施作业要求。