基于视觉注意机制的棉花污染物机器视觉检测算法

A machine vision inspection algorithm for contamination in cotton based on visual attention mechanism

针对棉花中污染物这类大背景中的目标检测,模仿人类视觉注意机制,提出了一种检测算法:在预注意阶段,主相机获取全局图像,利用离散余弦变换和支持向量机提取和识别特征,确定目标所在的感兴趣区域;在注意阶段,对应感兴趣区域的从相机工作,获取该区域的局部图像,利用均值和方差方法识别污染物。实验表明,该算法能去除冗余数据,提高检测精确度。

Simulating human visual attention mechanism, aiming at the real-time inspection for small target in a large background, such as contamination in cotton, an algorithm is proposed. In the re-attention stage, the host camera samples a global image, and locates the suspicious target region of interest in the image combining with discrete cosine transform （DCT） and support vector machine（SVM） for extracting and recognize of features; in the attention stage, the slave cameras sample the local images corresponding to the region respectively, and recognize the contamination in the local images through a method based on average and variance. The results indicated that the algorithm can be used for identification of target area in machine vision inspection, reduced the redundant data of the sampled images, and improved the accuracy of the machine vision system.