Capsule endoscopic system and position recognition method for detection picture of running area thereof

Abstract

The invention discloses a capsule endoscopic system and a position identification method for detecting pictures in a running area thereof. The position identification method comprises the steps of selecting image data; classifying the selected images, and adding markers to the classified images to represent their different positions; the current depth network model being obtained by pre-training and layer-by-layer adjustment of the position recognition network model. The training set is used to train the current depth network model. Whether the recognition accuracy, sensitivity and specificityof the current depth network model meet the preset requirements is judged. If so, the current network model after training is determined to be the depth network model after training. When the capsuleis used for inspection, the corresponding position of the image is obtained by using the trained depth network model. The invention can automatically identify the picture position collected by the magnetic control capsule, and provides a basis for determining whether the capsule completes the inspection of the whole running area.

Description

technical field [0001] The present invention relates to the technical field of medical equipment and image recognition, in particular to a capsule endoscope system and a method for recognizing the position of a picture in an operating area, which is used to automatically identify the position of a picture collected by a capsule, so as to determine whether the capsule has completed the entire operating area. Check to provide evidence. Background technique [0002] Due to individual differences in objects (especially inside organisms), how to identify the location of their internal structures has always been a research problem for those skilled in the art. [0003] In recent years, with the development of computer artificial intelligence and big data deep learning, new technological strength has been injected into the field of intelligent medicine. [0004] For example, in an application scenario of the present invention: when using capsule endoscopy to examine the digestive ...

AI Technical Summary

Problems solved by technology

However, this method adopts the traditional network model training, that is, randomly assigns the initial value of the network, which cannot satisfy the judgment of areas with fewer samples (such as cardia and gastric angle), and reduces the accuracy of judgment.

Images