Multi-target detection method and device and application thereof

Abstract

The invention provides a multi-target detection method and device and application thereof. The method comprises the steps of obtaining a to-be-detected image; inputting the to-be-detected image into a multi-target detection model to obtain a detection result; wherein the multi-target detection model comprises a backbone network, a neck module and a prediction head which are connected in sequence; each trunk layer is used for obtaining a second feature map with multi-scale feature information and after channel attention weighting; the neck module is used for performing feature aggregation on the second feature maps output by different trunk layers to obtain an aggregated feature map; the prediction head is used for performing multi-target detection according to the aggregated feature map. The method not only can extract multi-scale feature information, but also can flexibly adjust the parameters of the convolution kernel according to different sizes of the input to-be-detected image, more effectively extract finer-grained multi-scale space information, establish a longer-distance channel dependency relationship, and improve the detection accuracy. And adaptively carrying out feature calibration on the multi-scale channel attention vector again.

Description

technical field [0001] The present application relates to the field of target detection, in particular to a multi-target detection method, device and application thereof. Background technique [0002] In recent years, object detection algorithms have made great breakthroughs. At present, the mainstream target detection algorithms are divided into two categories according to the algorithm stage: the first category is the two-stage target detection algorithm, such as the R-CNN algorithm R-CNN in the Region Proposal, Fast R-CNN, Faster R-CNN, etc., They need to generate the target candidate frame by the algorithm, that is, the target position in the image, and then classify and regress the target candidate frame; the second category is single-stage target detection algorithms, such as Yolo and SSD algorithms, which only Use a convolutional neural network CNN to directly predict the categories and locations of different objects. The first type of method is more accurate, but t...

AI Technical Summary

Problems solved by technology

The advantages of spatial invariance include: variable parameters such as parameter sharing and translation, and its shortcomings are also obvious: the extracted features are relatively single, and the parameters of the convolution kernel cannot be flexibly adjusted according to the input target size; while the channel-specific convolution There is redundancy in the product kernel in the channel dimension

That is to say, on the one hand, conventional convolution deprives the convolution kernel of the ability to adapt to different visual patterns in different spatial positions, limits the receptive field of convolution, and is more difficult to detect small targets or blurred images; on the other hand, The inter-channel redundancy inside the convolution kernel limits the flexibility of the convolution kernel for different channels

Images