Early gastric cancer image recognition method based on evolutionary neural network model compression

Abstract

The invention discloses an early gastric cancer image recognition method based on evolutionary neural network model compression, relates to the technical field of computer application, and solves theproblems of parameter redundancy in an existing convolutional neural network model and poor real-time performance caused by large calculated amount consumption of the neural network model in an earlygastric cancer recognition process. The method comprises the following steps: collecting an early gastric cancer image data set; training a neural network model; constructing a binary encoding mode toencode parameters in the neural network model; compressing the trained neural network model by using an evolutionary algorithm; and finely adjusting the compressed neural network model, and identifying an early gastric cancer lesion area on a newly input gastroscopy image. According to the invention, the neural network model for detection can be compressed according to the collected gastroscopy image data, so that the efficiency of the neural network model in an early gastric cancer detection task is improved, and the deep neural network method has good real-time performance in the aspect ofearly gastric cancer detection.

Description

technical field [0001] The invention relates to the technical field of computer applications, in particular to an early gastric cancer image recognition method based on evolutionary neural network model compression. Background technique [0002] Gastric cancer (GC) is the third most lethal malignancy worldwide. Due to the mild early symptoms, gastric cancer is usually diagnosed at an advanced stage, and its 5-year survival rate is less than 30%. If gastric cancer is detected early and treated accordingly, its 5-year survival rate can increase to over 95%. Therefore, the identification of early gastric cancer is of great significance to reduce the mortality rate of gastric cancer. Gastroscopy has been widely used in the diagnosis of early gastric cancer, which can provide guidance for early intervention and treatment. Since early gastric cancer usually only shows some subtle changes in the mucosa, the sensitivity of gastric cancer identification is generally relatively low...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide an image recognition method for early gastric cancer based on evolutionary neural network model compression based on the deficiencies of the prior art. The method is based on an evolutionary algorithm and can efficiently and automatically discover Parameter redundancy, to solve the problem of poor real-time performance caused by the large computational consumption of the neural network model in the early gastric cancer recognition process

Images