US 11,967,071 B2
Method, device, apparatus, and medium for training recognition model and recognizing fundus features
Rui Wang, Guangdong (CN); Junhuan Zhao, Guangdong (CN); Lilong Wang, Guangdong (CN); Yuanzhi Yuan, Guangdong (CN); and Chuanfeng Lv, Guangdong (CN)
Assigned to PING AN TECHNOLOGY (SHENZHEN) CO., LTD., Shenzhen (CN)
Appl. No. 17/620,780
Filed by PING AN TECHNOLOGY (SHENZHEN) CO., LTD., Guangdong (CN)
PCT Filed Nov. 11, 2019, PCT No. PCT/CN2019/116940
§ 371(c)(1), (2) Date Dec. 20, 2021,
PCT Pub. No. WO2021/051519, PCT Pub. Date Mar. 25, 2021.
Claims priority of application No. 201910882247.3 (CN), filed on Sep. 18, 2019.
Prior Publication US 2022/0414868 A1, Dec. 29, 2022
Int. Cl. G06T 7/00 (2017.01); G06N 3/045 (2023.01); G06T 7/90 (2017.01); G06V 10/82 (2022.01)
CPC G06T 7/0012 (2013.01) [G06N 3/045 (2023.01); G06T 7/90 (2017.01); G06V 10/82 (2022.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30041 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A method for training recognition model, comprising:
obtaining a color fundus image sample associated with a label value, inputting the color fundus image sample into a preset recognition model containing initial parameters, wherein the preset recognition model comprises an input unit, a first convolutional neural network, and a second convolutional neural network sequentially connected together;
extracting a red channel image in a red channel from the color fundus image sample in the input unit;
inputting the red channel image into the first convolutional neural network to obtain a first recognition result and a feature image of the red channel image;
combining the color fundus image sample with the feature image to generate a combined image, and inputting the combined image into the second convolutional neural network to obtain a second recognition result;
inputting the label value, the first recognition result, and the second recognition result into a preset loss function to obtain a total loss value; wherein the loss function comprises a first loss weight of the first convolutional neural network and a second loss weight of the second convolutional neural network;
when the total loss value is less than or equal to a preset loss threshold, ending the training of the preset recognition model.