Anchor-frame-free target detection method based on lightweight convolution

Abstract

The invention relates to an anchor-frame-free target detection method based on lightweight convolution, and belongs to the field of computer vision target detection. The method comprises the following steps: S1, constructing a lightweight backbone network, inputting a picture into the lightweight backbone network, and extracting a feature map; S2, pooling an upper left corner point and a lower right corner point according to the obtained feature map; S3, performing cross star deformation convolution and angular point prediction operation on the pooled upper left corner point and the pooled lower right corner point respectively; and S4, carrying out angular point matching according to the predicted angular points and a centripetal displacement algorithm, and outputting a final result according to the score of the predicted bounding box. The anchor-frame-free design idea is adopted, the problems of imbalance of positive and negative samples, complex hyper-parameter design, complex calculation of anchor frames and the like can be solved in the training process, and meanwhile the lightweight backbone network can achieve the effects of effectively compressing the network model, reducing the model parameter quantity, reducing the calculation resource consumption and improving the calculation speed.

Description

technical field [0001] The invention belongs to the field of computer vision target detection, and relates to an anchor frame-free target detection method based on lightweight convolution. Background technique [0002] With the rapid development of deep learning, target detection has attracted more and more researchers' attention. Using anchor boxes to determine the position of target objects is a commonly used method in target detection. In recent years, the design idea of ​​no anchor frames has been introduced. The target detection of the anchor box puts aside the idea of ​​a large number of a priori candidate boxes, and directly classifies and predicts the location of the target object. Compared with the previous method based on anchor boxes, it has more advantages: reduce the use of hyperparameters; reduce the consumption of a large amount of memory resources; solve the problem of imbalance between positive and negative samples, etc. Therefore, object detection based on...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the purpose of the present invention is to provide an anchor-free target detection method based on lightweight convolution. The method adopts the design idea of ​​anchor-free, which can solve the imbalance of positive and negative samples in the training process, and the hyperparameter Complicated design, complex calculation of anchor frames, etc. At the same time, the use of lightweight backbone network can effectively compress the network model, reduce the amount of model parameters, reduce the consumption of computing resources, and improve the effect of computing speed

Images