基于跨连接卷积神经网络的绝缘子检测方法

针对电力设备图像中绝缘子所占比例较小和容易漏检的问题,提出了基于跨连接卷积神经网络的绝缘子检测方法。首先,通过在区域建议网络将网络后三层的卷积层分别和全连接层连接,使得这三层的卷积特征同时送入分类层和回归层,从而得到一系列高质量的绝缘子候选区域;将得到的候选区域映射绝缘子检测子网络,通过将得到的感兴趣区域特征送入级联的Adaboost分类器,实现对绝缘子的检测。对所提出跨连接卷积神经网络生成的候选区域进行了评估,并对不同的绝缘子检测方法进行了对比实验。实验结果表明,该方法得到的候选区域召回率高且更集中于绝缘子所在位置,绝缘子检测准确率比常规方法高出10%。所提方法能较好地对复杂背景图像中不同大小的绝缘子进行有效识别和精确定位。

Insulator Detection Method Based on Cross-connected Convolutional Neural Network

Aiming at the problem that the proportion of insulators in power equipment images is small and they are easy to miss detection, this paper proposes an insulator detection method based on cross-connected convolutional neural network for the power equipment image. Firstly, the convolutional layers of the last three layers of the network are connected with the fully-connected layer in the regional proposal network(RPN), and the three-layer convolution features are simultaneously sent to the classification layer and the regression layer to obtain a series of high quality insulator candidate regions. Secondly, the region proposals are input into the insulator detection sub-network, and the region of interest(ROI)features of candidate regions are sent to the cascaded Adaboost classifier to detect insulators. Evaluations are performed and comparative experiments are conducted based on the candidate region generation methods. The results show that the candidate regions obtained by the proposed method have high recall rates and they focus more on the positions of the insulators, and the accuracy of the insulator detection is 10% higher than the conventional methods. The proposed method can effectively recognize and locate insulators of different sizes with complex background.