Lightweight underwater target detection method based on feature fusion and neural network search

A technology of neural network and feature fusion, applied in the field of computer vision, to achieve the effect of promoting development

Pending Publication Date: 2021-12-31
DALIAN UNIV OF TECH
View PDF2 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] The present invention aims to solve the shortcomings of the existing deep learning-based underwater target detection method, and provides an underwater detection method based on a supervised learning-based land target detection method to assist underwater target detection network structure search

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
  • Lightweight underwater target detection method based on feature fusion and neural network search
  • Lightweight underwater target detection method based on feature fusion and neural network search
  • Lightweight underwater target detection method based on feature fusion and neural network search

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0033]The present invention searches and designs the underwater detection network based on the feature fusion operation and the gradient neural network search strategy. Under the premise of not introducing additional enhancement modules, loss functions or other tasks, the method proposed in this application can be used to design a fusion land The underwater detection network structure affected by the upper detection prior and the underwater quality factor, the search result example is as follows image 3 Shown, the specific implementation scheme of the present application is as follows:

[0034] The first step is to prepare the dataset. The pre-prepared datasets include pre-training datasets, land detection network training datasets and underwater detection network training datasets. The role of the pre-training data set is to pre-train the backbone of the land and underwater detection network, and obtain the pre-training weights to speed up the convergence process of the lat...

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 discloses a lightweight underwater target detection method based on feature fusion and neural network search, and belongs to the field of computer vision. Land detection network features and underwater detection network features are fused, namely addition operation is carried out, so that the goal of guiding underwater network structure construction by using land priori knowledge is achieved. Meanwhile, a neural network search algorithm is utilized, an efficient search space is designed, a gradient-based microsearchable strategy is adopted, and an underwater super-network structure and a land mirror image detection structure are constructed to directly establish a relation among underwater decline quality factors, land prior information and a detection network structure.

Description

technical field [0001] The invention belongs to the field of computer vision and relates to a lightweight underwater target detection method based on feature fusion and neural network search. Background technique [0002] Underwater object detection has a wide range of applications in oceanography, underwater navigation, and fishery. The research aims to locate the target of interest in the underwater image / video, and accurately determine the specific category of each target and the location and size of its bounding box. Compared with land-based detection scenarios, there are various underwater degradation problems such as sediment occlusion, color distortion, uneven illumination, and target occlusion. Therefore, underwater detection faces even greater challenges. In addition, underwater detection is usually a mobile terminal operation, which has high requirements for algorithm complexity. However, the current detection algorithms are often based on large-scale network str...

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
IPC IPC(8): G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06N3/045G06F18/241G06F18/253G06F18/214Y02A90/30
Inventor 刘日升付陈平仲维樊鑫罗钟铉
Owner DALIAN UNIV OF TECH
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

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap