Research On Target Classifition Through Scattering Media Based On Speckle Autocorrelation

As a carrier of target information,light has very important applications in many aspects,but in daily life,the intricate environment will affect our judgment and acquisition of target information.After analyzing and researching the various existing technologies for obtaining information from irregularly scattered light fields,in order to solve the shortcomings of poor adaptability and robustness of the existing target classification methods after the scattering medium,we choose the speckle automatic Related methods are used to process speckle and obtain the target object information hidden behind the scattering medium.In order to solve the problem of low efficiency of traditional feature extraction and classification algorithms,a convolutional neural network is selected to classify and recognize the processed speckle data.This project mainly completed the following tas ks:This paper analyzes and studies the current development and research status of the technology of hiding the target object after dealing with the scattering medium,and elaborates on it in detail.After understanding the advantages and disadvantages of various methods,the final scheme is selected and confirmed.Then the principle and application conditions of the speckle autocorrelation method,namely the optical memory effect,are analyzed theoretically,and the lens-like property of the scattering medium is also proved theoretically,which establishes the theoretical basis for future experiments.In order to get the experimental parameters of the final experiment,the experiment of the optical memory effect range of the scattering medium was designed and completed.The parameters needed for the final experiment could be confirmed by measuring the relative thickness of the scattering pla te and the optical memory effect range.The experiment of obtaining speckle image of target object was designed and completed.The experimental target was changed and a series of speckle images were obtained.After autocorrelation processing,the speckle i mages were divided into training set,verification set and training set.The appropriate convolutional neural network structure and parameters were selected to learn,train and test the data set obtained from the experiment.In order to improve its applicability,the system was tested under the two conditions of changing the scattering medium and changing the target state.The final experimental results show that speckle autocorrelation has a good effect on the information of the hidden target after process ing the scattering medium.Meanwhile,the convolutional neural network greatly improves the efficiency of obtaining the hidden target information and has a high recognition rate.Meanwhile,the adaptability and robustness of the system are greatly improved compared with the previous methods.