针对红外图像背景复杂, 分辨率低、对比度差等问题, 本文基于 RGB、HSV颜色空间转换和 Seam Carving缩放处理, 提出一种改进卷积神经网络(Convolutional Neural Network, CNN)的电力设备红外图像智能分类模型。首先, 着眼于 CNN的结构特点, 以 AlexNet网络模型为原型, 建立 CNN-Alex模型; 然后, 提出一种基于 RGB和 HSV颜色空间转换和基于 Seam Carving算法的设备红外图像处理方法, 分离目标设备红外背景及调整图像至合适大小, 对 CNN-Alex模型加以改进, 提高算法训练速度和准确率; 最后将改进 CNN模型与传统 BP模型和 CNN-Alex模型对比, 其训练集、验证集准确率分别为 99.5%、97.7%, 远优于对比模型, 验证了本文改进 CNN红外图像分类模型的良好适用性。

This paper proposes an improved CNN infrared image intelligent classification model for power equipment for complex background, low resolution, and poor contrast infrared images. The method is based on RGB and HSV color space conversion and seam carving scaling processing. First, the CNN-Alex model is built based on the AlexNet network model by focusing on the structural characteristics of CNN. Then, we propose an infrared image processing method based on RGB and HSV color space conversion and Seam Carving algorithm. The method separates the infrared background of the target device and adjusts the image size. It improves the CNN-Alex model by increasing the speed and accuracy of the algorithm. Finally, compared with the traditional BP model and the CNN-Alex model, the accuracy of the training and verification sets of the improved CNN model are 99.5% and 97.7%, which are far superior to the other models. This verifies the applicability of the improved CNN infrared image classification model.