Non-maximum suppression method, system and device based on attention mechanism and medium
A non-maximum value suppression and attention technology, applied in the field of image recognition, can solve the problems of target detection algorithms such as missing targets, low precision, and reduced detection accuracy, so as to improve accuracy, avoid missed detection, and reduce detection results the effect of the influence
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0043] The execution subject of the non-maximum value suppression method provided by the embodiment of the present invention may be a core processing unit in an image processing system, such as a GPU (Graphics Processing Unit, graphics processor). Wherein, a manner of implementing the non-maximum value suppression method provided in this embodiment may be a software or hardware circuit provided in the core processing unit. Of course, it should be emphasized that the execution body of the embodiment of the present invention is not limited to the above-mentioned core processing unit, and the manner of implementing the non-maximum value suppression method is not limited to the above-mentioned software or hardware circuit.
[0044] The target detection algorithm based on deep neural network only evaluates the quality of the candidate frame based on the confidence of the candidate frame, but in the target detection task, the important thing is to accurately frame the candidate frame...
Embodiment 2
[0083] like image 3 As shown, the second embodiment provides a non-maximum suppression system based on the attention mechanism based on the first embodiment, including:
[0084] A feature extraction unit 10, configured to obtain a target image containing a detection target, perform feature extraction on the target image, and obtain a feature map, wherein the feature map includes a plurality of candidate frames;
[0085] The first computing unit 20 is configured to perform weighted fusion processing on the candidate frame based on the attention mechanism algorithm, and obtain the first confidence value based on the attention mechanism in the candidate frame;
[0086] The second computing unit 30 is configured to preset a threshold of the full intersection and ratio loss function, and the weighted penalty algorithm based on the attention mechanism determines the first confidence corresponding to the candidate frame whose full intersection and ratio loss function value is greate...
Embodiment 3
[0090] like Figure 4 As shown, the third embodiment provides an electronic device, including: a processor 310 , a communication interface 320 , a memory 330 and a communication bus 340 , wherein the processor 310 , the communication interface 320 and the memory 330 communicate with each other through the communication bus 340 communication; the memory 330 is used to store computer programs; the processor 310 is used to execute the programs stored in the memory 330, so as to realize the non-polarity based attention mechanism described in any one of the first aspect. The large-value suppression method includes: step 110, obtaining a target image containing a detection target, performing feature extraction on the target image, and obtaining a feature map, wherein the feature map includes a plurality of candidate frames; step 120, based on an attention mechanism algorithm. Perform weighted fusion processing to obtain the first confidence value based on the attention mechanism in ...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More