A method for constructing a urine formed cell image classification model and a classification method
A classification model and construction method technology, applied in the field of image processing and deep convolutional neural network algorithm, to improve the speed of segmentation, reduce the difference, and increase the effect of diversity
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
specific Embodiment approach
[0059] The present invention expands the number of image units in the category, and the main purpose is to solve the problem of data imbalance, mainly by generating random number list, calculating remainder and random processing. The specific implementation is as follows:
[0060] First sort the original data samples from 0 to n-1 (n represents the total number of categories) according to the category order and generate the ID number of each type of image unit; then calculate the number of samples of each category, and record the category with the most samples quantity;
[0061] There are three categories shown in Table 1, the category numbers are 0, 1, and 2 respectively, and the ID number and image name of each category are shown in the table. It can be seen that the category with the largest number is category 2 with 5 pictures. For the convenience of identification, an image name is assigned to each image unit when the image is divided and saved.
[0062] Table 1
[00...
Embodiment 1
[0075] The original images were collected by the postgraduate laboratory of the Institute of Computer Vision and Network Intelligence, Xidian University. A total of 9624 pictures, each of which is processed in steps 1 and 2:
[0076] Step 1. Image preprocessing
[0077] (1) if figure 1 As shown, the original image size is 744*480, first reduce the image by about 3 times,
[0078] That is, the resize is 256*165;
[0079] (2) Using the multi-seed region growing algorithm, the image is binarized and segmented. In this algorithm, the number of seeds is set to 3, which are (64, 41), (192, 124) (128, 82), and the calculation method as follows:
[0080]
[0081]
[0082] (3) restore the segmented binarized image to the original size;
[0083] (4) Find the connected region, count the number of pixels in the connected region, if the number of pixels in the connected region is less than 500, discard the connected region to reduce the mixing of impurities;
[0084] (5) For ea...
Embodiment 2
[0099] In this embodiment, after step 2 of embodiment 1 is completed, the data expansion process is performed first, and then the image training set and the verification set are divided. The data distribution after processing is shown in Table 5 below.
[0100] table 5
[0101]
[0102]
[0103] As shown in Example 2 Table 5 Erythrocytes 1, the number of image units after the original segmentation is 2298. After the expansion process, the obtained number is 5562, and then the obtained number is obtained by randomly dividing the training set and the verification set at 7:3. There are 1000 original image units in the training set of 3893, accounting for 25.69%, and the image units after expansion processing account for 74.31%; in the verification set with the number of 1669, there are 1298 original image units, accounting for 33.3%. The images accounted for 66.7%.
[0104]Compared with the results of Example 1, as shown in Table 4 of Example 1, normal red blood cells 1, ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More