Research On Recognition Of The States Of Electrical Appliance Based On Infrared Image Processing

The health status of high-power heating and refrigeration appliances such as air conditioning directly affects electrical power consumption,and the condition monitoring for household appliances is helpful to prolong the service life of appliances and buildings’energy-saving.The distribution characteristics on temperature of such appliances are closely related to the working states,thermal infrared imaging technology is especially suitable for processing temperature data with significant variation distribution.Therefore,the state recognition methods of electrical appliance based on infrared image processing was researched in this thesis.The main research works are as follows:(1)Infrared image acquisition and pre-processing of working status of electrical appliances.Use the FOTRIC228 to take the the infrared video of the working state of household appliances,and use the AnalyzIR software to process the video to get the temperature matrix data and the infrared images of many synchronous frames separately.Three image enhancement methods are adopted such as gray transformation,histogram equalization and filtering,and Sobel operator is used for edge detection.(2)Propose a Recognition method for working states of electrical appliance based on temperature block’s features of infrared image.After pre-processing,the temperature matrix data of each frame of infrared image is overlapped and divided into blocks,and then the average value temperature of each block is extracted,take every three consecutive frames as a sample,and the average temperature and second-order difference values of each block of temperature matrix of each frame are extracted to form the characteristic matrix,and the similarity is used for classification.(3)Recognition for working states of electrical appliance based on deep learning.Mark the infrared images after pre-processing of infrared video by category sign.In Pytorch platform,six kinds of models,such as alexnet,googlenet,mobilenet,densenet,mnasnet and RESNET,are used to build the model,and then six kinds of final relatively optimized CNN models are obtained by verifying the samples.Take air conditioner,microwave oven,electric water bottle and LED as examples to verify the proposed method,the experimental results show that:(1)For the proposed method in this method,the average value of similarity between the test samples and the center sample of same category is above 0.8249 for each type of the four electrical appliances,and the one between the test samples and the center of other categories is less than 0.3005,the overall average recognition rate is 97.87%,the recognition rate of appliance’s working states based on deep learning is slightly higher than that of the proposed method,and the overall average recognition rate is 99.45%.(2)Compared with the time cost of the two methods,even for the lowest model of deep learning,the training time and test time are 1.78%and 0.45%of the proposed method respectively,but the hardware requirements are relatively high.It can be seen from the above results that the method proposed in this thesis has a certain practical value,and the project is worthy of further study.Figure[67]Table[20]Reference[86]...