Single-shot multi-frame detector optimization method based on bidirectional feature fusion and more balanced L1 loss

Abstract

The invention relates to a single-shot multi-frame detector optimization method based on bidirectional feature fusion and more balanced L1 loss, and belongs to the field of computer vision. The methodcomprises: A1, preprocessing training set images; a2, constructing a traditional SSD model; a3, modifying the traditional SSD algorithm model based on bidirectional feature fusion and a more balancedL1 loss function, and constructing a BFSSD model; a4, training the BFSSD algorithm model; and A5, testing the performance of the BFSSD algorithm model. According to the method, the problems of positive and negative samples and multi-task imbalance in the training process of a traditional SSD algorithm are solved, and the method has high robustness for small target detection.

Description

technical field [0001] The invention belongs to the field of computer vision and relates to a single-shot multi-frame detector optimization algorithm based on bidirectional feature fusion and more balanced L1 loss. Background technique [0002] Object detection is one of the core tasks of computer vision, which is widely used in intelligent monitoring, automatic driving and other fields. In recent years, various object detection methods based on Deep Convolutional Neural Network (DCNN) have achieved remarkable performance, improving the accuracy and speed of object detection. Object detection methods based on deep convolutional neural networks can be roughly divided into two categories: [0003] (1) Two-stage detection framework, which first generates a series of target candidate regions, and then extracts the features of the target candidate regions through a deep neural network, and uses these features for classification and target ground-truth bounding box regression. T...

AI Technical Summary

Problems solved by technology

[0006] In view of this, the purpose of the present invention is to provide a single-shot multi-frame detector optimization method based on two-way feature fusion and more balanced L1 loss, for the traditional SSD algorithm does not consider the difference between different feature layers due to the independent use of multi-scale feature layers relationship, and then ignore some context information, resulting in the problem of not being robust enough for small target detection, as well as the problem of positive and negative samples and multi-task imbalance in the training process of the traditional SSD algorithm.

Images