Text detection model training method and device, readable storage medium and equipment

Abstract

The embodiment of the invention provides a text detection model training method and device, a readable storage medium and equipment. The training method comprises the following steps: inputting a to-be-processed sample image into a convolutional network model to obtain a predicted value; obtaining an annotation value of the sample image; obtaining predicted loss according to the annotation value, the predicted value and the loss function; adjusting parameters of the convolutional network model according to the prediction loss, wherein the loss function comprises a simple sample judgment function and a weight coefficient function; the simple sample judgment function is used for filtering the sample images with the prediction confidence greater than a preset first threshold and the sample images with the prediction confidence less than a preset second threshold, and the weight coefficient function is used for adjusting the weight of the unfiltered sample images. According to the embodiment of the invention, a simple positive sample and a simple negative sample can be filtered, and the model can pay attention to a more valuable sample image in combination with the adjustment of the weight value.

Description

technical field [0001] The invention relates to the technical field of text detection model training, in particular to a text detection model training method, device, readable storage medium and equipment. Background technique [0002] In intelligent education scenarios, the positioning of image text areas is the pre-process for text recognition and content understanding, and the detection accuracy of text lines directly affects the processing effect of subsequent tasks. At present, text detection models based on deep learning are divided into two categories: regression methods based on preset boxes and pixel segmentation methods based on text regions. Among them, the pixel segmentation method based on the text area has strong adaptability, and has obvious advantages for slender text and curved text. In the text detection model based on the pixel segmentation method, the text segmentation task divides the image into text areas and non-text areas, which is a typical binary c...

AI Technical Summary

Problems solved by technology

However, the division standard of difficult and easy samples of the FocalLoss method relies too much on the confidence of the model prediction. When a training sample is mislabeled, the adaptive weighting strategy of the FocalLoss method has a large weight difference between different samples, which is likely to cause unstable training. process, and even bring about the problem of model training divergence

Especially for the text detection task of the segmentation method, there is no clear texture boundary between the text area and the non-text area, and there must be a lot of noise in the labeling process. The FocalLoss method that relies on confidence to define difficult and easy samples is difficult to play a role.

Images