基于数字图像的棉田复杂背景下棉蚜统计方法

A Statistical Method for Counting Cotton Aphis under Complex Background in Cotton Field Based on Digital Image

【目的】实现棉田复杂背景下棉蚜快速准确计数,提出一种先彩色分割,后自适应构元素及阈值的棉蚜计数方法。【方法】该方法基于大量棉蚜图像RGB数据进行K-means聚类建模,利用结构元素完成腐蚀去噪,针对黏连区域像素个数进行求模运算。【结果】根据图像颜色特征将噪音分为13类,蚜虫分为7类,得到其RGB值后再次分类,并分析数据建立模型实现蚜虫和噪音的彩色分割;根据统计学原理建立结构元素,对不同噪音的图像自动选择最优结构元素进行腐蚀去噪;计算黏连区域像素个数与单头蚜虫期望大小像素个数的模,实现黏连区域蚜虫计数。【结论】基于结构元素的棉蚜计数方法能有效的对棉田复杂背景下棉蚜快速准确计数,计数平均准确率为86.47%,在图像处理过程中极大降低了算法对阈值的依赖性,有效地解决了棉蚜图像黏连分割的问题,完成基于数字图像的复杂背景下棉蚜计数。

【Objective】 This paper aims to present a new automatic counting method for cotton aphids in the hope of achieving the rapid and accurate counting of aphids in cotton fields under complex backgrounds.【Method】A large amount of RGB data of cotton aphids was analyzed by a K-means clustering algorithm to obtain an accurate model. Autonomous structural elements were used to complete the corrosion de-noising, and a modulo operation was performed on the number of pixels in the overlapping area. First, the noises were divided into 13 categories according to the colors of images, and the aphids were divided into 7 types. Then, the aphids were classified again after the RGB data of each type was obtained. The data were then analyzed to establish models for the color segmentation of aphids and noises. Next, the association of autonomous structural elements was established according to the principle of statistics, and the optimal structure elements of the images with different noise levels were selected for corrosion de-noising. Finally, the number of cotton aphids in the overlapping area was counted by performing a modulo operation based on the number of pixels of the overlapping area and the expected size of the single-headed aphids.【Result】Experimental results showed that the method proposed in this paper can effectively and accurately count cotton aphids in cotton fields under a complex background with an average accuracy of 86.47%. Besides this, in the process of image processing, the dependence of the algorithm on threshold was greatly reduced and the problem of image adhesion segmentation of cotton aphid was solved effectively. Finally, the count of cotton aphid in complex background based on digital image was completed.【Conclusion】.