A deep learning training sample optimization method

Abstract

The invention discloses a deep learning training sample optimization method, comprising the following steps: drawing contour element labeling information of an original image, stripping the contour element labeling information of the original image, generating a single contour element original sub-image and overlapping contour element original sub-image, image enhancement processing, and edge detection processing to obtain the original sub-image of a single edge contour element and the original sub-image of an edge-overlapping contour element, as well as pooling processing and image segmentation processing; the present invention can ensure that the original image has a higher The image quality is the basis for improving the stability of post-sequence optimization processing. By performing edge detection processing on the image after image enhancement processing, the sample element boundary error in the original image of the deep learning training sample can be eliminated, and the image quality and training samples can be improved. The generation efficiency can effectively shorten the training time of deep learning training samples.

Description

technical field [0001] The invention relates to the technical field of data processing, in particular to a deep learning training sample optimization method. Background technique [0002] In many application scenarios, in order to achieve the accuracy of object detection, when using the deep learning model for training, the deep learning model needs to learn and train a large number of high-quality input data sets. This data set is generally large enough, representative and It has relatively clear labels, but deep learning has strong learning ability or fitting ability. The more complex the network model, the stronger its ability, and the more training data is needed, otherwise it is easy to produce overfitting phenomenon. [0003] At present, in the deep learning training sample optimization method, the sample element boundary error in the original image is relatively large, which affects the generation efficiency and quality of the deep learning training sample. As a resu...

AI Technical Summary

Problems solved by technology

[0002] In many application scenarios, in order to achieve the accuracy of object detection, when using the deep learning model for training, the deep learning model needs to learn and train a large number of high-quality input data sets. This data set is generally large enough, representative and It has relatively clear labels, but deep learning has strong learning ability or fitting ability. The more complex the network model, the stronger its ability, and more training data is needed, otherwise it is easy to produce overfitting phenomenon

[0003] At present, in the deep learning training sample optimization method, the sample element boundary error in the original image is relatively large, which affects the generation efficiency and quality of the deep learning training sample. As a result, the training results are not accurate and stable enough, and the training results of deep learning training samples will also be unstable. Therefore, the present invention proposes a deep learning training sample optimization method to solve the deficiencies in the prior art

Images