基于改进的慢特征分析模型的目标变化监测方法研究

Abstract: With the popularization of public transportation, the security problem has been paid much attention to, and it is very important to maintain and maintain the equipment regularly. However, most of the equipment maintenance and repair is mainly dependent on the related technical staff, and some devices need to be maintained are not easy to detect and have a large number. Therefore, this project will use artificial intelligence to study the maintenance and maintenance of important devices in some commonly used devices. The main research contents include:.(1) how to locate and detect the video target in the video captured by the complex industrial environment. By optimizing and comparing the traditional image target detection methods, the location and detection based on the video target are obtained..(2) how to set up the target image sequence to get the image feature of the long time change of the target. By extracting the single target frame image in video, the single target frame image is registered and aligned to form the time sequence of the single target frame image..(3) how to use the target image sequence to analyze the slow change of the target, and to predict and distinguish the next change of the target. The time series image is modeled by slow feature analysis model, and the relationship between feature and physical characteristics of target surface is obtained, and it is identified and predicted by LSTM neural network..This project is of great economic and social value for improving the maintenance of intelligent devices and ensuring industrial transformation and upgrading in China.

随着公共交通方式的普及，其安全性问题也受到广泛的重视，对设备的定期检修及维护显得至关重要。然而目前大多数器件的维护和检修难以进行检测且数量众多，因此本项目将利用人工智能研究某些常用设备中重要器件的检修和维护。主要研究内容包括： .（1）在复杂工业环境所拍摄的视频中，如何实现对视频目标的定位与检测。通过对传统图像目标检测的方法进行优化和对比，定位和检测视频目标。.（2）如何建立目标图像序列，以得到目标长时间变化的图像特征。通过提取视频中单目标帧图像，将单目标帧图像配准和对齐，形成单目标帧图像的时间序列。.（3）如何利用目标图像序列分析出目标的缓慢变化，并预测和判别目标的下次变化。对慢特征模型进行改进，利用改进的慢特征分析模型对时间序列图像进行建模，得到特征与目标表面物理特性的联系，并通过LSTM神经网络进行判别和预测。.本项目对于提升我国智能器件维护以及保障工业转型升级具有重要价值。

随着公共交通方式的普及，其安全性问题也受到广泛的重视，对设备的定期检修及维护显得至关重要。然而目前大多数器件的维护和检修难以进行检测且数量众多，因此本项目将利用人工智能研究某些常用设备中重要器件的检修和维护。主要研究内容包括： .（1）在复杂工业环境所拍摄的视频中，如何实现对视频目标的定位与检测。通过对传统图像目标检测的方法进行优化和对比，定位和检测视频目标。.（2）如何建立目标图像序列，以得到目标长时间变化的图像特征。通过提取视频中单目标帧图像，将单目标帧图像配准和对齐，形成单目标帧图像的时间序列。.（3）如何利用目标图像序列分析出目标的缓慢变化，并预测和判别目标的下次变化。对慢特征模型进行改进，利用改进的慢特征分析模型对时间序列图像进行建模，得到特征与目标表面物理特性的联系，并通过LSTM神经网络进行判别和预测。.本项目对于提升我国智能器件维护以及保障工业转型升级具有重要价值。