A Pattern Recognition Method Based on Deep Convolutional Neural Network

Abstract

The invention discloses a pattern recognition method based on a deep convolutional neural network, mainly a pattern recognition method based on a deep convolutional neural network and introducing an attention mechanism and image enhancement means. First get the sample data set, and the relevant professionals will mark the image of the data set and do grayscale processing on the image, which can highlight the outline of the main target, which is conducive to feature extraction, and then use random rotation to expand the data set at different angles , then enhance and preprocess the image data, and finally construct a deep convolutional neural network that can efficiently extract image features, use the established data set to perform 5-fold cross-validation to train and test the model, and complete the visual aided detection model build. The invention has higher operation efficiency in sample identification, reduces model parameters, reduces resource occupation, and has high demand for software and hardware, and can be better put into practical use.

Description

technical field [0001] The invention belongs to the field of deep learning computer vision, and mainly relates to a pattern recognition method based on a deep convolutional neural network and introducing an attention mechanism and image enhancement means. Background technique [0002] "Feature extraction + classifier" is a classic framework in the field of pattern recognition, that is, to represent images by constructing features manually, and then send the image data at the feature level to the classifier to realize the classification and recognition of target images. Neurological research shows that the human brain does not extract features in the process of processing visual images, but transmits signals to a deep network composed of a large number of neurons and passes them layer by layer to finally obtain the implicit expression of the signal. Deep learning is to let the image propagate in the network and output an effective representation of the image by simulating the...

AI Technical Summary

Problems solved by technology

However, these existing methods still have some defects

First of all, when labeling and segmenting images, operators need to have professional knowledge in related fields and be able to make professional interpretations of labeling sample points or texture features, and the features extracted by different methods may be different, and some images are very complex. If the computer can automatically extract the features in the image and classify the image, the result may be more objective and the classification accuracy will be higher

Images