基于HOG局部双线性插值的机械零部件检测与识别

Mechanical Parts Detection and Recognition Based on Local Bilinear Interpolation of HOG

传统方向梯度直方图(HOG)特征方法用于机械零部件检测时,对零部件的微小缺陷点的检测和识别不是很有效,因此课题组提出了一种局部双线性插值的HOG特征提取改进算法,将其用于小缺陷点的零件的检测和识别。首先,在图像内部的某个块上实现局部双线性插值;然后对局部双线性插值图像块进行梯度计算,提取新的梯度方向直方图;最后利用神经网络分类方法对具有微小缺陷点的零件进行检测。实验结果表明:局部双线性插值HOG特征提取方法比传统的HOG特征提取方法具有更好的检测性能;增强图像的抗混叠识别效果。

The traditional HOG feature method in the detection of mechanical parts are not very effective for the detection and recognition of small defect points of industrial parts.Therefore,an improved HOG feature extraction algorithm based on local bilinear interpolation was proposed for the detection and identification of mechanical parts which have small defects.Firstly,local bilinear interpolation was implemented on some internal block inside the image.Then the gradient of the local bilinear interpolation image block was calculated and the histogram of the new oriented gradient were abstracted.Finally,the components with tiny flaw point by neural network classification were detected.The experimental results demonstrate that the local bilinear interpolation HOG features extraction offers better detection performance than the conventional HOG feature extraction method and it can also enhance the effect of anti-aliasing image recognition.