Image classification method and related device

Abstract

The invention discloses an image classification method, which comprises the following steps executed by an image classification model obtained by carrying out model training by utilizing a cross entropy function: receiving a target image, and carrying out quantization conversion on the target image; acting on the quantized and converted target image through a convolution kernel with a preset number of channels to obtain a preset number of feature images; performing convolution operation on the feature image, and performing product pooling on convolution output after convolution operation through a 2 * 2 pooling window; and after product pooling, obtaining a score value of the target image for each target type on an output layer, calculating a probability value of the target image belongingto each target type based on each score value by utilizing a softmax function, and determining the target type with the maximum probability value as the type to which the target image belongs. The method can effectively improve the accuracy and reliability of image classification. The invention further discloses an image classification device and equipment and a computer readable storage medium,which all have the above technical effects.

Description

technical field [0001] The present application relates to the technical field of artificial intelligence, and in particular to an image classification method; it also relates to an image classification device, equipment, and computer-readable storage medium. Background technique [0002] Convolutional neural network has the characteristics of local perception, parameter sharing and multi-core convolution. It is widely used in the field of image recognition, including face recognition, pedestrian detection, speech recognition, medical image processing, etc., even if the direction, shape, and lighting conditions of the target object And so on, it still has good classification accuracy. Quantum computing adopts a new type of computing method that is completely different from traditional computing methods. Quantum parallel processing greatly improves the efficiency of quantum computing, enabling it to achieve a problem-solving speed that cannot be achieved by conventional comput...

AI Technical Summary

Problems solved by technology

However, the current quantum gate circuit neural network has the following shortcomings: First, in the pooling process, the product of two adjacent positions is used to achieve the purpose of pooling. This asymmetric sampling on two-dimensional data makes Most of the quantum gate circuit neural network is in an asymmetric state during the operation process, and at the same time it needs to face the problem of selecting the sampling direction; the second is the loss function selection of the output layer, which uses the square of the difference between the predicted value and the target value, resulting in convergence The speed is slow; the above two technical defects will affect the accuracy and reliability of image classification

Images