基于深度学习的工程作业智能监控技术的模型优化测试

Model optimization test of intellegent monitoringtechnology for engineering operation

为了改善输变电工程作业人工检测效率低、人员财产安全风险频发问题,提出了一种基于神经网络模型压缩目标检测技术,结合改进的卡兹曼滤波进行目标跟踪,实现无人机对输变电工程的智能监控。通过深度可分离卷积模型,降低参数数量、提高推理速度,引入注意力机制和剪枝算法,降低神经模型复杂度和非必要参数,减少信息处理数量。同时采用欧氏距离改进的卡尔曼滤波进行目标跟踪,提升目标跟踪的实时性和准确性。实验结果表明,提出的组合模型算法与传统算法对比,可以充分提取文本的高频特征信息,对于不同颗粒度的数据集的准确率提升8%,召回率降低4%,性能更优,具有一定的科研及应用价值。

In order to improve the low efficiency of manual detection and frequent risks of personnel and property safety in power transmission and transformation projects,a compressed target detection technology based on neural network model wasproposed inthis paper.The technology was combined with improved kazman filter for target tracking,to achieve intelligent monitoring of power transmission and transformation projects by UAVs.Firstly,the deep separable convolution model was used to reduce the number of parameters and improve the reasoning speed,and the attention mechanism and pruning algorithm were introduced to reduce the complexity and unnecessary parameters of the neural model and reduce the amount of information processing.At the same time,the improved Kalman filter using Euclidean distance was used for target tracking to improve the real-time and accuracy of target tracking.The results showed that,compared with the traditional algorithm,the proposed combinatorial model algorithm fully extracted the high-frequency feature information of the text,and the accuracy of data sets with different granularity was increased by 8%,the recall rate was reduced by 4%,and the performance was better.In conclusion,theproposed combinatorial model algorithm has certain scientific research and application value.