基于支持向量机的镁熔液弱小目标检测

Detection of Dim Target in Magnesium Alloy Melt Based on Support Vector Machine

针对镁熔液图像中弱小目标不易检测的问题,提出一种基于支持向量机回归(SVR)的第一气泡检测方法。首先利用支持向量机回归原理的函数回归特性对原图像进行背景逼近;再重构两帧残差图像并对其进行帧差运算;最后通过所提出的阈值分割方法处理并利用形态学开运算识别第一气泡。与BP神经网络背景预测算法对比,SVR算法所获取的帧差图像在信噪比和信噪比增益方面分别提高了17.67%和17.69%,且在实时性方面较优。

For difficult detection of dim target in magnesium alloy melt images,a detection method based on support vector machine regression（SVR）was proposed.At first,the function approximation characteristic of SVR theory was utilized to estimate background of original images.Then the two reconstructed images were obtained and the operation of frame difference was used for the two reconstructed images.At last,the dim target was recognized by using threshold segmentation proposed and morphological opening operation.Compared with algorithm of BP neural network background prediction,the frame difference image obtained in algorithm proposed is improved by 17.67% and by 17.69%,respectively,in signal to noise ratio（SNR）and SNR gain,and the algorithm also exhibits an advantage in real-time.