CAJ | 학술논문

针对传统方法存在目标特征提取不足和计算复杂度高等问题，从人脑视觉认知出发，提出一种仿人脑视皮层机制的目标识别方法。引入卷积神经网络，提出基于 Gabor卷积核的目标边缘特征，以拟合简单细胞感受野；结合初级视皮层腹侧通路目标识别方式，模拟视皮层信息处理机制，抽取边缘图像 PHOG特征来描述目标，建立 PHOG特征的仿人脑皮层识别模型。采用多类 SVM分类器对特征向量进行识别实验，实验结果表明，该方法减少了计算复杂度，提高了识别率。
침대전통방법존재목표특정제취불족화계산복잡도고등문제，종인뇌시각인지출발，제출일충방인뇌시피층궤제적목표식별방법。인입권적신경망락，제출기우 Gabor권적핵적목표변연특정，이의합간단세포감수야；결합초급시피층복측통로목표식별방식，모의시피층신식처리궤제，추취변연도상 PHOG특정래묘술목표，건립 PHOG특정적방인뇌피층식별모형。채용다류 SVM분류기대특정향량진행식별실험，실험결과표명，해방법감소료계산복잡도，제고료식별솔。
Aiming at the shortage of feature extraction and high computation complexity in traditional methods,a method was proposed for target recognition imitating human brain visual cortex-like mechanisms inspired from the human visual perception. A convolutional neural network was introduced and the target edge characteristics based on Gabor convolution kernels were put forward to fit the simple cell’s receptive field.Combined with the obj ect recognition mode in primary visual cortex ventral path-way,a human brain visual cortex-like mechanism was simulated to extract the PHOG (pyramid histogram of oriented gradient) feature and a PHOG-based hierarchical model was established via imitating human brain visual cortical.A multi-class SVM clas-sifier was used to classify the feature vectors.Results of experiments show that the proposed method not only simplifies the cal-culation,but also improves the recognition accuracy.