Method for optimizing convolutional neural network based on linear decoder and interpolation sampling

Abstract

The invention belongs to the field of image recognition, and particularly relates to a method for optimizing a convolutional neural network based on a linear decoder and interpolation sampling. The constructed convolutional neural network comprises an input layer, a convolution layer, a pooling layer, a full-connection network layer and an output layer. The method comprises the following steps: firstly, training with a convolution linear encoder to obtain a weight value, and taking the weight value as an initial value of the convolution neural network, then performing multi-interpolation sampling and pooling on the characteristic pattern obtained by the convolution layer, and through the forward propagation and reverse adjustment, finally obtaining the local gradient of each neuron in eachlayer and the weight value of the convolution kernel of each convolution layer. Through the comparison with an existing method, the experiment result shows that when the convolutional neural networkconstructed by the method is used for classifying images, the convolutional neural network has the advantages of being higher in convergence speed and higher in accuracy.

Description

technical field [0001] The invention belongs to the field of image recognition, in particular to a method for optimizing a convolutional neural network based on a linear decoder and interpolation sampling. Image recognition refers to the technology of using computers to process, analyze and understand images to identify targets and objects in various patterns. It is the basis for practical technologies such as stereo vision, motion analysis, and data fusion. Background technique [0002] The linear decoder is a special neural network, similar to the self-encoder neural network, which tries to approximate an identity function, so that the output consists of 3 layers of neurons, namely the input layer, the hidden layer and the output layer. For the hidden layer, neurons still use the S-type (or tanh) activation function, but for the output layer, the output uses the identity activation function as the activation function. Because the output range of the S-type activation fun...

AI Technical Summary

Problems solved by technology

Because the output range of the S-type activation function is [0,1], when the output layer uses the activation function, it is necessary to limit or scale the input so that it is in the range of [0,1]. Some data sets, such as MINST, can It is very convenient to scale the output to [0,1], but it is difficult to meet the requirements for the input, but the above problems can be easily solved when using the identity activation function

Images