MB-SSD method and MB-SSD feature extraction network suitable for target detection

Abstract

The invention discloses an MB-SSD method suitable for target detection. The method comprises the steps: extracting a plurality of small target images, and carrying out the enhancement of the extractedplurality of small target images through employing a generative adversarial network; constructing an MB-SSD feature extraction network, the MB-SSD feature extraction network comprising a main branchfeature extraction network, a branch feature extraction network and a positioning network, inputting the enhanced data into the MB-SSD feature extraction network, and respectively obtaining classification positioning results of the main branch feature extraction network and the branch feature extraction network; adjusting the output characteristics of the classification network according to the IoU overlap ratio of different candidate boxes in the same area on the positioning network; fusing the classification positioning results of the main branch feature extraction network and the branch feature extraction network, and performing dimension reduction; calculating model loss, training the model, and optimizing model parameters. According to the method, a method for adjusting the classification result by adding the relative overlap ratio into the classification layer can be added to improve the classification effect of the SSD algorithm, Meanwhile, the detection precision of the small target is effectively improved.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to an MB-SSD method suitable for target detection and an MB-SSD feature extraction network. Background technique [0002] Object detection is a popular research direction in the field of computer vision, which can be applied to unmanned driving, video surveillance, pedestrian detection, remote sensing image detection and other fields. Traditional target detection algorithms first extract features manually, such as SIFT (Scale Invariant Feature Transform), HOG (Histogram of Oriented Gradient), SURF (Speeded Up RobustFeatures) etc., and then combine these artificially extracted features with the classifier for target recognition, and finally combine the corresponding strategy to locate the position of the target, but for scenes with complex and changeable backgrounds and targets, it is difficult for people to summarize the image. Abstract features, so traditional methods have...

AI Technical Summary

Problems solved by technology

For the detection of large targets in sparse scenes, the detection accuracy of the SDD algorithm can exceed that of the more accurate Faster R-CNN, but the detection effect for small targets is not good, and there are cases of missing and wrong detection of small targets.

Images