基于深度学习的弱纹理图像关键目标点识别定位方法

Method for identifying and locating key target points in weak texture images based on deep learning

摘要:

为提高弱纹理图像关键目标点的检测识别能力，提出基于深度学习的弱纹理图像关键目标点识别定位方法。构建低光照强度弱纹理图像关键目标点的拓扑特征分布模型，采用透射率作为检测系数，结合亮通道的先验知识，建立像素大数据分布集，采用暗原色融合和RGB像素分解方法实现对低光照强度弱纹理图像的信息自适应增强处理；根据透射区域噪点融合匹配结果，采用交叉组合滤波检测和深度学习算法，实现对低光照强度弱纹理图像降噪和信息增强，据此实现对低光照强度弱纹理图像关键目标点检测识别。仿真结果表明，采用该方法定位识别的精度较高，平均为0.93，图像输出质量较好，峰值信噪比平均为32.87，通过准确率-召回率曲线的对比也表明性能较为优越。

Abstract:

In order to improve the detection and recognition ability of key target points in weak texture images, a method of identifying and locating key target points in weak texture images based on deep learning is proposed. The topological feature distribution model of key target points of weak texture images with low light intensity is constructed. The transmittance is used as the detection coefficient, and the pixel big data distribution set is established by combining the prior knowledge of bright channels. The dark primary color fusion and RGB pixel decomposition methods are used to realize the information adaptive enhancement processing of weak texture images with low light intensity. According to the results of noise fusion and matching in the transmission area, the cross combination filter detection and deep learning algorithm are adopted to realize the noise reduction and information enhancement of the weak texture image with low light intensity, thus realizing the key target point detection and recognition of the weak texture image with low light intensity. Simulation results show that this method has high positioning recognition accuracy, with an average of 0.93, and good image output quality, with an average peak signal-to-noise ratio of 32.87. By comparing the accuracy-recall curve, the performance of this method is superior.