Adaptive feature and data distribution target detection method

Abstract

The invention discloses an adaptive feature and data distribution target detection method, and relates to the technical field of computer vision. The invention aims to provide a target detection method for self-adaptive characteristics and data distribution, which can automatically select certain parameters of certain layers as a characteristic pattern to perform target detection, effectively reduce redundant parameters and save the time for optimizing a network. The method comprises the following steps of: 1) inputting an image by adopting a pixel with the width w of 416, a pixel with the height h of 416 and a channel number c of 3; the method comprises the steps of (1) channel weight calculation, (2) size reduction and channel number increase module operation, (3) two-path dense connection module operation, (4) feature map generation, (5) learning, calculation and re-screening of weights of all channels, and (6) regression of target positions and categories.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a Dense Attention Network object detection method for adaptive features and data distribution. Background technique [0002] In the field of computer vision, convolutional neural networks are widely used. Especially after AlexNet won the championship in the ImageNet competition in 2012 with an advantage of 10.9 percentage points over the second place, convolutional neural networks have been widely used in the field of computer vision, and AlexNet has also laid a milestone foundation for deep learning. [0003] Karen Simonyan and Andrew Zisserman from Oxford University proposed VGG-Nets in 2014, and achieved excellent results in the first place in the positioning task and the second place in the classification task in the 2014 ImageNet competition. VGG-Nets show that the performance of the network can be improved by increasing the number of network layers and depth based ...

AI Technical Summary

Problems solved by technology

[0010] Although the existing network structure is getting deeper and wider, it has achieved a good effect on some public data sets, but it also brings more parameters

Since it is impossible to determine which layer or which parameters play a key role, it is also impossible to determine whether a certain layer or certain parameters are actually used, which to some extent causes the redundancy of parameters

[0011] 2. It is difficult to optimize for actual tasks

[0012] The design of existing networks is aimed at more complex data sets. These data sets have many categories, but there may not be so many categories in actual tasks, the data is not necessarily particularly complex, and the size distribution of the targets is relatively relatively large. concentrated

While ensuring the accuracy of the network, in order to reduce the computing resources of the network, the usual practice is to reduce some layers, but it is impossible to determine which layers to reduce are effective enough

It must be compared through a large number of experiments, which will consume a lot of time and energy

[0013] 3. The network training phase cannot adapt to the distribution of data

[0014] Most of the existing networks are designed with multiple feature pyramids FPN, but the training loss weight of each FPN is the same

The size distribution of targets in actual tasks is often relatively concentrated, and most of the detection results may come from one FPN, so it is obviously unscientific that the loss weights of each FPN are exactly the same

Images