Cell image detection and segmentation method for generating candidate anchor boxes based on clustering

Abstract

The invention discloses a cell image detection and segmentation method for generating candidate anchor boxes based on clustering. The cell image detection and segmentation method comprises the following steps: step 1, making a data set; step 2, data set sample dimension feature statistics: setting ISODATA clustering algorithm initial parameters, performing statistics on sample dimension information through a clustering algorithm, and generating a sample dimension proportion; step 3, cell feature extraction and fusion, including the following steps: 3.1, building a feature extraction network; 3.2, performing feature multi-scale fusion; step 4, generating a cancer cell target area candidate box, and sending the fused features and the target sample dimension proportion into an RPN network togenerate a target area; step 5, refining a detection target result of the cancer cell image; and step 6, segmentation Mask generation of the cancer cell image. According to the cell image detection and segmentation method, the generated candidate anchor boxes are enabled to better fit a real sample dimension rule; the difficulty of candidate box regression is reduced; the algorithm regression speed is improved; and the detection and segmentation performance is improved.

Description

technical field [0001] The invention belongs to the field of computer vision and discloses a cell image detection and segmentation method for generating candidate anchor frames based on clustering. Specifically, the clustering algorithm is used to count the dimensions of the samples in the data set, and the dimension conversion ratio is sent to the deep learning Mask R-CNN algorithm as the aspect ratio of the candidate anchor frame in the RPN network to realize the detection and segmentation of cancer cell images. The ISODATA clustering algorithm makes statistics on the sample dimensions, and then converts the statistical results into the dimension ratio as the parameter setting of the aspect ratio of the candidate anchor box in the RPN network, and finally realizes the detection and segmentation of cancer cell images through the Mask R-CNN algorithm. Background technique [0002] Research at the molecular and cellular level is an important step in the development of new dru...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the problem that the aspect ratio of candidate anchor boxes in the existing deep learning two-stage algorithm cannot better fit the distribution of sample dimensions in the data set and meet the characteristics of large differences in cancer cell size, the present invention provides a method using a clustering algorithm to generate A cell image detection and segmentation method with a candidate anchor frame aspect ratio that is more in line with the real sample dimension distribution

Images