基于区域生长法和BP神经网络的红外图像识别

针对变电站巡检机器人远程监控系统中红外图像识别存在的问题,提出一种基于改进区域生长法和BP神经网络的红外图像目标设备分割与识别的方法。利用最小二乘法拟合出红外图像中亮度与温度之间的线性关系,建立基于像素的图像温度场；根据设定温度范围确定区域生长法的种子点位置,利用Otsu法确定截屏窗口最优分割阈值,并结合灰度相似性阈值作为区域生长法的分割准则,实现该窗口目标设备精确分割；将分割出的设备二值图像的Hu不变矩作为设备形状特征向量,并对其进行不变性和类间区分度验证；采用引入附加动量法和自适应调整学习率的BP神经网络实现多种电气设备的识别,实验数据表明优化后的BP神经网络具有迭代收敛快,误差波动性小,分类准确度高等特点。

Infrared image recognition based on region growing method and BP neural network

Aiming at the problems in the process of infrared image identification in the remote monitoring system of substation inspection robot,a segmentation and recognition method of infrared image target equipment based on improved region growing method and BP neural network is proposed.The linear relationship between the brightness and the temperature in the infrared image is fitted by the least squares method,and the temperature field is established based on the pixel.The seed point position of the region growing method is determined according to the set temperature range.The Otsu method is used to determine the optimal segmentation threshold,and the gray level similarity threshold is adopted as the segmentation criterion of the region growing method to complete the optimal segmentation of the window target area.The Hu invariant moments of the binary image of the device are considered as the feature vector of the device shape,and its invariant and interclass divisions are verified.The recognition of the electrical equipment is realized by the BP neural network introducing the additional momentum method and the adaptive adjustment learning rate,and the experimental data show that the optimized BP neural network has the characteristics of fast iteration,poor error volatility,and high classification accuracy.