抽水蓄能电站烟火智能识别算法的改进研究

Research on the Improvement of Intelligent Recognition Algorithm for Pumped-storage Hydroelectricity Fireworks

提出一种基于级联深度神经网络的抽水蓄能电站烟火智能识别算法,通过使用级联多监督卷积神经网络对图像进行去噪,可解决图像背景噪声干扰大的问题;研究提出利用级联深度神经网络将优化YOLO-v5作为一级网络对烟火目标进行定位和识别,可初步预估烟火特征和位置;将Inception-v3网络作为二级网络进行更具针对性识别和训练烟火特征,可进一步提高识别准确率。实验结果表明,该算法比YOLO-v5和Inception-v3具有更高的准确率,能够满足复杂场景检测的需要,有效提高抽水蓄能电站复杂场景中检测准确率和精准度。

This study proposes an intelligent recognition algorithm for pumped-storage hydroelectricity fireworks based on a cascade depth neural network,which uses cascade multi-supervised convolutional neural network to denoise images to solve the problem of large background noise interference.A new method is proposed,which uses the optimized YOLO-v5 as the primary network to locate and identify the pyrotechnic target,and can predict the feature and position of the pyrotechnic;Inception-v3 network is used as a secondary network to identify and train firework features more pertinently,which can further improve the accuracy of identification.The experimental results show that the proposed algorithm has higher accuracy than YOLO-v5 and Inception-v3,and can meet the needs of complex scene detection,and effectively improve the accuracy and precision of pumped-storage hydroelectricity complex scene detection.