Research On Garbage Discarding In Reservoir Scene

With the development of computer technology and information technology,human behavior recognition has become a popular research direction in the field of computer vision.Image classification is an important basic problem in computer vision.,it is also the foundation of other high-level visual tasks such as behavior recognition and object tracking.Convolution neural network--CNN,which as a deep learning algorithm,shows well in image classification tasks by its own unique features such as parameter sharing,local connection.The problem of garbage discarding in the reservoir scene is an important part of the reservoir environment management.In order to solve this problem,this paper divided the task into two stages: first stage,distinguished whether the object in the two intercepted frame carry garbage or not.Second stage,discriminated whether the object who carry garbage in the last frame is the same person who don't carry garbage in the next frame.If yes,it could be considered that the object who carried garbage in the previous frame discarded the garbage.In view of the task in first stage,after discussed the advantages of convolution neural network in solving the problem of image classification,this paper proposed a recognition method based on Alex Net model under the migration learning model for small scale training samples,and compared the CNN-with_dropout model and the CNN-without_dropout model.The experimental results showed that the convolution neural network model under the migration learning model has achieved good results in the first stage of garbage discarding in the reservoir scene,and the dropout technology can effectively reduce the degree of overfitting of the model.For the second stage task,this paper proposed to use the image similarity to determine whether it is the same object,and compared three kinds of Hash algorithm:mean hash,difference hash,perception hash to calculate.Finally it found that the mean hash algorithm is more suitable for the task requirement than the other two algorithms because of it's more strict requirements and it pays more attention to local details.In order to evaluate the accuracy of the algorithm,the images of the same object who carry garbage and don't carry garbage was taken as an example.The mean hash algorithm was used tocalculate the similarity of two images to 85.37%,and the result is reasonable.Therefore,the experimental results showed that the two phase method: CNN-with_dropout model and the mean hash,which proposed in this paper can solve the problem of garbage discarding in the reservoir scene.