Image recognition method and device

Abstract

The invention provides an image recognition method and device. The method comprises the steps that an image is acquired; pixels of interest in the image are marked according to the pixel values of pixels within the image; according to the distribution of the pixels of interest in the image, noise in the pixels of interest is excluded; a machine learning model is used to recognize the image markedwith the remaining pixels of interest after exclusion; and the pixels of a recognition target are marked in the image, wherein the machine learning model is acquired by training a sample image markedwith the pixels of the recognition target and a sample image marked with the pixels of a non-recognition target.

Description

technical field [0001] The invention relates to the field of medical image processing, in particular to an image recognition method and equipment. Background technique [0002] Using machine learning algorithms and models to recognize images is an efficient way, and it is also the underlying technology in many fields such as autonomous driving, smart cameras, and robots. [0003] Before using a machine learning model (such as a neural network) for image recognition, the model must first be trained using sample images. The training method is usually: 1. Manually mark the target of interest in the image and generate target area identification information; 2. Input the identification information and images generated by annotation into the deep neural network; 3. Train the deep neural network and wait for it to converge. The trained machine learning model can then be used to identify and label objects of interest from the image. [0004] For the training phase of the model, in...

AI Technical Summary

Problems solved by technology

[0004] For the training phase of the model, in the above method, the first step requires the annotator to draw all possible target areas in the picture. For the undrawn targets, the machine learning model will default to the background, affecting the training results, and then Affect the recognition accuracy

The limitations of the existing image recognition methods based on machine learning are: a large number of labeled pictures are required, and the labor cost is high. labeling is a big problem

Therefore, only a single and obvious target of interest is marked in most sample images

[0005] For the output results of the model, since the features in the sample images used in the training stage are relatively simple, when the machine learning model is used to directly recognize the image, for complex images, the uncertainty of the position, shape, and size of the region of interest is very large. Strong, which makes the output of the model less accurate

Images