受神经解剖学和心理学中有关视觉系统研究成果的启发，提出一种基于生物视觉仿生机理的铜带表面缺陷检测模型。该模型首先使用Gaussian金字塔分解和Gabor滤波器提取缺陷图像特征，合成特征显著图，然后模拟自底向上注意机制，分析视网膜中央凹的内容以获取what信息；再根据扫视仿真中访问点的时间顺序序列，即扫描路径，组成where信息流；最后利用离散的可观测马尔可夫模型，根据what信息和where信息调整相应类的单个马尔可夫链的概率，最大化训练样本的似然值，从而实现缺陷的正确分类。实验结果表明，本文算法在表面缺陷检测系统中的可行性及有效性，在多类缺陷分类中达到94.40%的总准确率。

An inspection model for surface defects of copper strips based on vision bionics is proposed in this paper. The model firstly uses Gaussian pyramid and Gabor filter to extract the features of defective images, and combines the features into a saliency map. Then, it simulates the bottom-up attention mechanism, analyzing the content of fovea to acquire “what” information and obtaining a time-ordered sequence of visited regions after simulation of saccades, which constitutes the “where” stream. At last, it uses a discrete and observable Markov model to adjust the probabilities of a single Markov chain according to “what” and “where” stream, and maximizes the likelihood of the training data, and realizes correct classification. The experimental results demonstrate the feasibility and effectiveness of the proposed method in surface defect inspection, furthermore, the average accuracy rate can reach 94.40%.