基于改进的ResNet50网络的光伏热斑识别算法

Photovoltaic Hot Spot Recognition Algorithm Based on Improved ResNet50 Network

为提高样本在分布不均衡情况下的识别精度,提出一种改进的ResNet50卷积神经网络光伏热斑识别算法。首先,为增加初期红外纹理信息流入、调整网络宽度,设计一种头部分组特征提取模块,并将其嵌入到残差网络中,提高网络在图像细微特征方面的提取能力;然后,将通道注意力机制与残差模块相结合,增加网络通道间的热斑特征信息权重,提高模型识别性能和网络收敛速度;最后,通过图像转换HSV颜色空间、平均H分量梯度直方图峰值等数据预处理方法,将负样本转为多分类数据集,并用于热斑识别网络模型,实现热斑识别结果的可视化。实验结果表明,对比其他算法,改进后的ResNet50网络在识别精度上得到显著提高。

To improve the recognition accuracy of samples under uneven distribution,an improved ResNet50 convolutional neural network photovoltaic hot spot recognition algorithm is proposed.First,in order to increase the initial infrared texture information inflow and adjust the network width,a head grouping feature extraction module is designed and embedded into the residual network to improve the network’s extraction capability of image subtle features;Then,by combining channel attention mechanism with residual module,the weight of hot spot feature information between network channels is increased to improve model recognition performance and network convergence speed;Last,through data preprocessing methods such as image conversion into HSV color space and average H-component gradient histogram peak,negative samples are converted into multi classification datasets and used in the hot spot recognition network model to achieve visualization of the hot spot recognition results.Results show that compared to other algorithms,improved ResNet50 network significantly improves detection accuracy.