Quasi-circle target detection method based on convolutional neural network and MAML algorithm

Abstract

The invention discloses a quasi-circle target detection method based on a convolutional neural network and an MAML algorithm. The method comprises the following steps: firstly, collecting and marking a quasi-circle target image data set; dividing the collected data set so as to carry out different subtasks; performing data enhancement on images in a training set; using a large target detection data set BO, a small target detection data set SO, a high-resolution detection data set HR and a low-resolution detection data set LR to train a large target detection model, a small target detection model, a high-resolution detection model and a low-resolution detection model respectively; learning a most potential quasi-circle detection model by using an MAML algorithm; and sending different test samples into the trained model for reasoning to obtain a detection result.

Description

technical field [0001] The invention relates to the technical fields of computer vision and pattern recognition, in particular to a circle-like object detection method based on a convolutional neural network and a MAML algorithm. Background technique [0002] Object detection is an important research direction in the field of computer vision, among which circle-like object detection can promote the rapid development of ball game adjudication, face tracking and other tasks. At present, with the wide application of neural networks in the field of computer vision, object detection methods based on convolutional neural networks have also emerged. However, most of the existing target detection methods use a rotatable rectangular frame as the labeling method for target detection. For each target, a large number of candidate frames will be generated. Each candidate frame will at least include the center position, width and height, and rotation. The angle has 5 parameters, which le...

AI Technical Summary

Problems solved by technology

However, most of the existing target detection methods use a rotatable rectangular frame as the labeling method for target detection. For each target, a large number of candidate frames will be generated. Each candidate frame will at least include the center position, width and height, and rotation. Angle has 5 parameters, which leads to a very large amount of calculation, so it consumes a lot of computing resources

Images