Focus image sample determination method based on artificial intelligence and related device

Abstract

The embodiment of the invention discloses a focus image sample determination method based on artificial intelligence. When a network model used for recognizing a focus image is trained, a training sample determined according to the focus image is adopted, and the training frequency of the training sample and loss parameters determined according to the first i-1 rounds of training can be determinedbefore the network model is subjected to the i-th round of training. The sampling weight of the training sample corresponding to the ith round of training is determined according to the loss parameter and the training times. Therefore, the target sample required by the i-th round of training is determined by adopting the weight; the determined target sample is not difficult to be too simple; in each round of training, the sampling weight of each training sample is dynamically adjusted according to historical training information; therefore, the quality of the training sample determined by each round of training is improved, excessive repeated training of the simple sample is avoided, a foundation is laid for the model training quality, and the recognition precision of the network model for the focus image is remarkably improved.

Description

technical field [0001] The present application relates to the field of data processing, in particular to an artificial intelligence-based method for determining lesion image samples and related devices. Background technique [0002] With the development of artificial intelligence, object detection, such as lesion recognition, can be quickly performed on images through network models. For some network models that need to be trained by training samples before they can be put into use, in order to ensure the detection accuracy of the model, it is very important to determine a reasonable training sample. [0003] However, in some lesion recognition scenarios, the training samples that can be obtained have obvious defects, for example, most of the training samples are relatively simple, and only a small part of the training samples are relatively difficult. For example, in the identification of pulmonary nodule lesions, the area where the lesion is located in the lesion image is...

AI Technical Summary

Problems solved by technology

[0004] In this case, if a reasonable training sample cannot be determined during the training process, the model will repeatedly learn simple samples during most of the training time, making it difficult to accurately identify the lesion in the lesion image after training

Images