Convolutional neural network model for static image behavior recognition

Abstract

The invention discloses a convolutional neural network model for static image behavior recognition, and relates to the technical field of image processing methods. The model comprises a VGG16 convolution layer module used for processing an input image; a multi-branch convolution layer module used for respectively learning features output by the last layer of the VGG16 convolution layer module through different branches, wherein the convolution layer weight of the module is randomly initial in the training process; and a softmax classifier layer module used for classifying the features outputby the three-branch convolution layer module. The model can explore information in a VGG16 convolutional layer module channel more effectively, and can play a role in finely adjusting the highest layer weight in the neural network; therefore, the action information in the single static human behavior image can be identified more effectively.

Description

technical field

[0001] The invention relates to the technical field of image processing methods, in particular to a convolutional neural network model for static image behavior recognition. Background technique

[0002] Human action recognition is one of the important research contents in the field of computer vision. Most of the research on human behavior recognition is based on video rather than single image methods, but there are many common human behaviors that can be fully represented by a single image, for example, phone calls, computer interaction, shooting Wait. Even if the video information of these actions is available, methods based on static cues are still required, such as playing guitar, riding a horse, running, etc., that is to say, the range of motion of these human actions is small, and the movement trajectory is not discriminative, so it is still necessary to recognize these actions. Static method based on a single image.

[0003] Algorithms based on dee...

Images