Research On Object Tracking Based On Deep Learning And Application In Leucorrhea Recognition

With the advent of the era of big data,object tracking has received more and more attention and development in computer vision research,and it has also been applied more and more in people’s lives and industrial production.Although the object tracking research has made great progress and breakthroughs in recent years,due to the complexity of the external environment and the influence of noise factors such as the deformation of the object itself,achieving a robust object tracking algorithm has always been a the task of great challenge.The core problem of object tracking is feature expression.The early features are selected manually,and with the different application scenarios,objected selection of appropriate features is required.Even so,the tracking effect is far from satisfying the actual application requirements.In recent years,with the study of deep learning and the successes achieved,the researchers who tracked the object have found new ideas.Compared with the manually selected features and shallow features,the feature extraction ability and strong migration ability through deep learning are utilized,to build a feature extraction module for the object tracking algorithm.This paper studies the object tracking algorithm based on deep learning and deeply analyzes all aspects of the object tracking algorithm.The current mainstream method of object tracking does not require large samples as a support.We have attempted to use this feature to solve the detection of similar objects(such as mold,red blood cells,and white blood cells)in clinical medical images.This is an important starting point for this study.Innovation.Through a large number of experiments,we successfully applied the new method of object tracking in our study to a specific clinical medical image object detection.The main research content of this article is as follows:(1)This thesis firstly analyzes and elaborates the related theories of object tracking,aiming at the problems of high computational complexity,large training parameters,long time-consuming and large sample data requirements for feature extraction framework based on traditional deep learning.This paper attempts a new deep learning network model PCANet network to extract features of the object.The network performs PCA filtering and dimension reduction on the extracted features,and then uses binary hash encoding as the nonlinear output,and the last pooling.The layer is reduced by the block histogram method to obtain the feature vector of the object image.(2)Around the object tracking algorithm,it is easy to produce the problem of losing object under the interference of complex external circumstances.In this paper,EdgeBoxes is used to generate candidate objects and combined with the detector localization method,and a robust algorithm is implemented.Firstly,the EdgeBoxes are used to extract the possible positions of the object,and then these detectors are used to classify and detect these regions.Finally,the object tracking is achieved.A threshold can be set to determine whether or not to update the network.The experimental results show that the tracking algorithm shows a good tracking effect when the object is moving fast or the object is partially occluded.(3)If traditional deep learning is used to achieve medical image recognition,we need a large number of samples.In order to avoid this problem,this paper finally uses the idea and method of object tracking to solve the problem of image recognition in leucomicroscopy.Artificial detection of mycotic vaginitis is easy to produce false detection and other problems.Combined with the development needs of intelligent detection and identification,a fungal detection algorithm based on PCANet network is proposed.After extracting the characteristics of mold by PCANet network,the classifier is trained,and the sample to be detected is used to generate a high-quality object candidate area through the selective search method.After being classified by the trained classifier,the object is located.Experiments show that this article can accurately detect mold in the complex environment of mold activity.