Research On Criminal Image Classification And Retrieval Based On Machine Learning

With the rapid development of our country's information technology,a large number of criminal investigation images have been collected for the detection of modern cases.These images not only contain many valuable clues,but also provide strong evidence,but due to the large number of images and the characteristics of the images themselves that makes it difficult to dig out this information,the traditional way that use artificial means to make use of images has been unable to meet the needs of modern criminal investigation cases.It is urgent to use modern information technology to achieve criminal investigation images automatic classification and image information mining,etc.,thereby improving management efficiency,reflecting the important value of criminal investigation image.Deep learning is an emerging technology developed in the field of machine learning in recent years.By simulating the process of cognitive things in the human brain,that has made many achievements for digging out image information,which traditional techniques can't compare,this article has learned a lot about this technology and research results,and combined with the classification and retrieval of criminal investigation images.The main research results are as follows:(1)In terms of image classification,this paper applies deep learning technology to criminal investigation images,and implements automatic classification and labeling.Research on a series of region-based convolutional neural network classification algorithms,and improves the Faster Rcnn named YF-Faster Rcnn.Using the forensic image obtained from the cooperation unit,a criminal investigation image data set was made.A large number of experiments were conducted on the data set.A large number of parameters were adjusted to tune the network,and multiple objects in the image could be identified.Using the improved method in this paper,the correct rate of classification results has been significantly improved.(2)In the aspect of image retrieval,this paper builds its own convolutional neural network,joins the pyramid pooling layer,performs network training on the training set produced in this paper,describes a deep learning feature YF-DL extraction method and uses the extracted deep learning features for image retrieval.This paper compares several traditional shallow features with Deep-learning features,and it was found that the retrieval performance using deep learning features was better than traditional features.This paper combines deep learning features with traditional features and studies the impact of multiple features on search results.Finally,combining semantic-based and content-based retrieval methods,the recall rate and precision rate are significantly improved compared to traditional retrieval.