Chip counting method based on deep learning

Abstract

The invention discloses a chip counting method based on deep learning. The chip counting method comprises: collecting chip images needing to be counted in various scenes; inputting a chip image into the backbone network, and through deep learning of image features, generating different scale features, and fusing the different scale features to obtain a feature map; sending the feature map into a confidence predictor to obtain a confidence map; sending the feature map obtained by multiplying the confidence map by the feature map to a threshold value encoder, and obtaining a threshold value map; sending the threshold map and the confidence map into a binarization layer, and filtering a chip region through a threshold to obtain a connected region segmentation map; and detecting the connected regions through the segmentation map, and counting the chips to obtain the final number of the chips. According to the method, huge data is utilized, model training is completed through deep learning, different images are mapped through adaptive learning threshold values, each chip is detected more accurately, and the method is more easily applied to chip counting of an actual scene and is not limited by the scene.

Description

technical field [0001] The invention relates to the field of machine vision counting, in particular to a deep learning-based chip counting method. Background technique [0002] In the production line, it is necessary to count the number of chips produced. Since there are various types of chips and the environments in which chips are produced are also different, it is a difficult problem to count different chips in different production environments. The commonly used method generally uses an inductive chip counting device. The installation scene of this device is strictly limited, and the counting scene is relatively single. The other is counting by human eyes, which requires extremely high labor and time costs. [0003] In the field of machine vision, traditional chip counting algorithms are only suitable for situations with obvious features and simple backgrounds. For example, the invention patent with the patent number CN109166116A discloses "an image processing algorit...

AI Technical Summary

Problems solved by technology

[0006] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is that the current chip counting method has poor robustness and low counting efficiency

Images