Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2020-09-28 Cooperative journals: 《计算机应用研究》
Abstract： Based on convolution neural network, this paper proposed a new asymmetric two branch underwater biological classification model to solve the problems of low visibility, poor illumination conditions and no obvious differences among species in the underwater environment. In the model, the interactive branch used different convolution neural network to extracted local features and interacted with local features through the interactive module to enhanced the classification model. Convolutional neural network branch could effectively learned the global characteristics of the target and made up for the global information ignored in the interactive branch. Finally, this model obtains 98.9%, 98.3% and 97.9% of the accuracy on the three data sets of fish4 knowledge (f4k) , Eilat and RAMAS, which are significantly improved compared with the previous methods. visual interpretation also verifies that the model can effectively capture local features and eliminates the background influence. Finally, it shows that the model has good classification performance in underwater environment.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-24 Cooperative journals: 《计算机应用研究》
Abstract： Auto-Encoder can express the semantic features of data through deep unsupervised learning, but it is hard to determine the nodes of hidden layer and the processing of data for classification often leads to low accuracy and low stability. To solve the problems, this paper proposes a semi-supervised auto-encoder using sparse and label regularizations (LSRAE) to extract the essential features of the samples more accurately by combining unsupervised learning with supervised learning. The sparse regularization term adds constraints to the response of each hidden node, so that this algorithm can find potential structures in the data when the number of hidden neurons is large. At the same time, this algorithm introduce a label regularization term to compare the actual labels with desired labels by supervised learning to adjust the network parameters and further improve the classification accuracy. In order to verify the validity of the proposed method, this algorithm tests many data sets in the experiment. The results show that compared with traditional auto-encoders (AE) , sparse auto-encoder (SAE) , and extreme learning machine (ELM) , SLRAE can obviously improve the classification accuracy and stability when the processed data is applied to the same classifier.
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-20 Cooperative journals: 《计算机应用研究》
Abstract： Due to the difference of handwriting caused by the large differences in edge contour, this paper proposed an algorithm named Jacobian regularized sparse automatic encoding machine (JSAE) for handwriting identification. This algorithm added sparse constraint and Jacobi regular item into the automatic coding machine, which improves the recognition accuracy. The sparse constraint can extract hidden structure from the data effectively and the regularized Jacobi can describe the marginal features of point data, thus it enables the learning ability of auto-encoder algorithm to improve and obtain the essential characteristics of the sample more accurately. Experimental results show that JSAE outperforms the basic auto-encoders (AE) and sparse auto-encoders(SAE) .
Subjects: Computer Science >> Integration Theory of Computer Science submitted time 2018-05-02 Cooperative journals: 《计算机应用研究》
Abstract： Aiming at the problem that the deep belief network (DBN) is susceptible to the training parameters燿uring the fine-tune process, this paper proposed a kind of batch normalization DBN classification method (BNDBN) . Firstly, this method used unsupervised learning to obtain high-level representation of raw data. Then through the introduction of scale transformation and translation transformation parameters, it processed the output characteristics of each layer by batch normalization. And it fed the post-processing characteristics into the nonlinear transformation activation layer. Finally, it trained and studied the parameters of the affine transformation and the original network by using the stochastic gradient descent method. The BNDBN method reduced the dependence of the gradient on the parameter size, which effectively resolved the problem of changing the value distribution of activation function caused by the change of network parameters and improves the training efficiency. To verify the effectiveness of the proposed method, it selected MNIST handwritten database and the USPS handwritten digital identification library for testing. Compared with the Dropout-DBN, DBN, ANN, SVM and KNN, the results show that the proposed method significantly improved the classification accuracy and had stronger feature extraction ability.
Hits
Hits
Downloads
Comment