Supercharge Your Innovation With Domain-Expert AI Agents!

Image saliency target detection method based on deep supervised learning

A technology of supervised learning and target detection, applied in neural learning methods, instruments, biological neural network models, etc., can solve problems such as large network architecture, consumption of computing resources, and difficulty in detecting prominent objects, and achieve clear boundaries and highlighted areas Uniform, promoting effect of complementary effect

Pending Publication Date: 2021-07-20
HANGZHOU DIANZI UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The VGG network architecture is small and has fewer parameters. The salient target detection model based on the VGG network is suitable as a preprocessing process for major visual tasks. At the same time, because of the light and small network, it is difficult for VGG to extract deep semantic information; Compared with the VGG network, the ResNet network has better performance, but the network architecture is very large, which consumes too much computing resources
Most other saliency detection models generate deep features by sequentially stacking convolutional layers and maximum pooling layers. These models mainly focus on the nonlinear combination of high-level features extracted from the last convolutional layer, and lack low-level visual information such as object edges. , it is difficult to detect salient objects in scenes with transparent objects, similar contrast between foreground and background, and complex backgrounds

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
  • Image saliency target detection method based on deep supervised learning
  • Image saliency target detection method based on deep supervised learning
  • Image saliency target detection method based on deep supervised learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0035] The specific embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings.

[0036] The present invention is an image saliency target detection method based on deep supervised learning, which solves the problems of boundary blur and the like in multi-scale saliency detection on the market. First, modify the VGG-16 network to adapt to the saliency detection task, remove the last pooling layer and all fully connected layers of the network, use the modified VGG-16 network to extract the multi-scale feature information of the image, and recursively fuse the multi-scale feature to get a saliency image. In order to strengthen the boundary of the image, the ground-truth image is sequentially down-sampled to the same size as the feature image, and the information from the pixel level supervises the saliency image prediction of each layer, promoting the complementary effect in the prediction, and recursively guiding...

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 an image saliency target detection method based on deep supervised learning, and the method comprises the steps of modifying a VGG-16 network to adapt to a saliency detection task, removing the last pooling layer and all full-connection layers of the network, extracting the multi-scale feature information of an image through the modified VGG-16 network, carrying out the recursive fusion of the multi-scale features, and obtaining a saliency image; in order to enhance the boundary of the image, down-sampling the truth value image to the same size as the feature image in sequence, using information from the pixel level to supervise saliency image prediction of each layer, promoting the complementary effect in prediction, guiding the saliency feature image of each layer recursively, optimizing boundary information, and enhancing the final saliency image effect. According to the invention, the problems of boundary blur and the like existing in multi-scale saliency detection in the prior art are solved.

Description

technical field [0001] The invention belongs to the field of image salient target detection, in particular to an image salient target detection method based on deep supervised learning. Background technique [0002] The purpose of salient target detection is to use the algorithm to locate the most obvious and eye-catching area in the image (that is, the area of ​​interest to the human eye), to reflect the importance of each area of ​​the image in the human eye, to identify the main body of the image, and to reduce the complexity of the scene. Researchers are committed to developing a computational model that simulates the human attention process to predict image saliency targets, which can be used as a preprocessing step for many computer vision tasks, such as scene classification, image segmentation, video compression, information hiding, etc. Crucial role. [0003] Over the past two decades, a large number of methods have been proposed to detect salient objects in images....

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): G06K9/00G06K9/46G06K9/62G06N3/04G06N3/08
CPCG06N3/08G06V20/46G06V20/41G06V10/462G06N3/045G06F18/253Y02T10/40
Inventor 张善卿孟一恒李黎陆剑锋
Owner HANGZHOU DIANZI UNIV
Features
  • R&D
  • Intellectual Property
  • Life Sciences
  • Materials
  • Tech Scout
Why Patsnap Eureka
  • Unparalleled Data Quality
  • Higher Quality Content
  • 60% Fewer Hallucinations
Social media
Patsnap Eureka Blog
Learn More