Target detection and identification method based on FSAF and fast-slow weight

A target detection and recognition method technology, applied in character and pattern recognition, instruments, biological neural network models, etc., can solve problems such as low average accuracy, improve accuracy and convergence speed, reduce computational complexity, and improve training accuracy Effect

Pending Publication Date: 2021-12-28
BEIJING INSTITUTE OF TECHNOLOGYGY
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0013] The purpose of the present invention is to propose a target detection and recognition method based on FSAF and fast-slow weights for the existing RetinaNet and FSAF-based target detection methods with low average accuracy.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Target detection and identification method based on FSAF and fast-slow weight
  • Target detection and identification method based on FSAF and fast-slow weight
  • Target detection and identification method based on FSAF and fast-slow weight

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0092] The present invention realizes the optimal selection of the feature layer based on the feature pyramid and RetinaNet network, combined with the fast-slow weight optimizer, there is a wide application space in underwater target recognition and target detection with fuzzy features in dark environments; in the marine industry, It has great practical significance in the application scenarios of fishery and land fuzzy environment. In the example, the simulated environment test is carried out on the training and testing data sets provided by the global underwater robot competition. Its data set is obtained from intercepting non-adjacent frames of videos in a simulated environment, and it has a good simulation of underwater blurred environments. In addition, it has achieved good results in classification and recognition data sets such as ImageNet, PASCAL VOC, and Labelme. Effect.

[0093] The configuration of the network model training and testing equipment used in the exampl...

Embodiment 2

[0186] This embodiment describes the specific implementation of the target detection and recognition method based on FSAF and fast-slow weights of the present invention, and realizes the performance comparison between the method and the RetinaNet network in the environment where the small data set is relatively clear, and shows two methods The comparison results of , including the following steps:

[0187] Step A. Divide the atlas used for comparing model effects into a training set with a number of 200 and an evaluation set with a number of 100, and the division method adopts random division;

[0188] During the specific implementation, the pictures come from the training and testing data sets provided by the underwater robot competition, and the data sets are obtained from the non-adjacent frames of the intercepted video in the simulated environment;

[0189] Step B, preprocessing the image;

[0190] Wherein graphics and processing are in order to obtain the image input mod...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention relates to a target detection and identification method based on FSAF and fast-slow weight, and belongs to the technical field of supervised learning and target identification. The method comprises the steps: 1) constructing a backbone network comprising a convolutional layer, a feature map layer and a prediction layer and a RetinaNet with a reference frame branch; 2) building an FSAF branch and generating an effective area and a neglected area of an image feature layer, specifically, adding a reference frame-free branch to a prediction layer of each level in a RetinaNet reference frame branch, reducing a mapping frame of a standard frame in different feature layers according to an A1-time proportion to serve as the effective area, and reducing the mapping frame according to an A2-time proportion to serve as the neglected area; 3) calculating the sum of classification loss and regression loss based on the FSAF branch, namely comprehensive loss; and 4) inputting the comprehensive loss corresponding to the optimal feature layer into a standard optimizer and a LookAhead optimizer, so as to converge the comprehensive loss. According to the method, a reference frame-free mechanism and a Lookahead fast-slow weight optimizer are introduced, so that a better recognition effect than a RetinaNet network is realized, and the identification accuracy and the loss convergence speed are both improved.

Description

technical field [0001] The invention relates to a target detection and recognition method based on FSAF and fast-slow weight, and belongs to the technical field of supervised learning and target recognition. Background technique [0002] With the continuous deepening and development of deep-sea observation and marine technology, the importance of underwater target recognition in fishery, aquaculture, coastal defense, and military applications has become increasingly prominent. However, the underwater environment in the natural environment is very complex, and various types of interference emerge in endlessly, which have a great impact on the effect of underwater target detection. Therefore, technologies focused on improving the accuracy of underwater target detection have also developed rapidly. [0003] From the perspective of target recognition, underwater image data has the following problems: (1) edge and detail degradation; (2) target and background contrast is extreme...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/32G06K9/62G06N3/04
CPCG06N3/048G06N3/045G06F18/253
Inventor 聂振钢赵乐卢继华侯杰继马志峰韩航程谢民
Owner BEIJING INSTITUTE OF TECHNOLOGYGY
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products