Road water body detection method combining reflection attention mechanism and self-attention mechanism
A technology that combines reflection and detection methods, applied in neural learning methods, computer components, character and pattern recognition, etc., can solve the problems of complex GAN training process, difficulty in convergence, and inability to improve stable performance, and achieve the goal of improving detection accuracy Effect
Active Publication Date: 2021-04-23
NANJING UNIV OF SCI & TECH
View PDF3 Cites 1 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
This method introduces GAN in image segmentation, which is a beneficial attempt of reinforcement learning in the field of image segmentation. However, the training process of GAN is complicated, it is difficult to converge, and stable performance cannot be improved.
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 moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0038] A road water body detection method combining reflective attention and self-attention mechanism, the specific steps are:
[0039] Step 1: Use a visible light camera to collect road images in different scenarios and scale them to a specified size to form a training set, label the water body area information in the training set images, and obtain the corresponding binary mask. Specifically, when annotating an image, an area with a pixel value of 0 indicates that the corresponding original image is a non-water body area, and an area with a pixel value of 255 indicates that the corresponding original image is a water body area. In some embodiments, the constructed deep neural network requires an input size of 360×640, so both the original road image and the binary mask are scaled to 360×640.
[0040]In a further embodiment, a data set input module is constructed. Specifically, first define a DataProvider class, which has two different working modes, corresponding to the roo...
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
Login to View More Abstract
The invention discloses a road water body detection method combining a reflection attention mechanism and a self-attention mechanism. The method comprises the steps of using a visible light camera for collecting a road water body image and labeling the road water body image; constructing a network model described in the invention, and training the network by using the training image and the annotation data; zooming an image to be detected to a specified size, and inputting the image to be detected into the trained network to obtain a binary image used for representing a water body region and a non-water body region. According to the method, water body detection can be efficiently carried out by utilizing the pavement images acquired by the visible light camera, the detection result is relatively fine and has good performance on various indexes, and the method is suitable for surface classification related tasks on unmanned vehicles.
Description
technical field [0001] The invention belongs to the image segmentation technology in computer vision, and specifically relates to a road water detection method combining reflective attention and self-attention mechanism. Background technique [0002] For water region detection in road images, most of the existing deep learning methods treat it as an image segmentation problem, while U-Net, as the baseline in the field of medical image segmentation, utilizes image features from both bottom and high layers to ensure The recovered segmentation results are finer. At the same time, the parameters of the network are not too large to prevent overfitting. However, due to the inconsistency in the nature of medical images and road water images, the lesion area in medical images is mostly small, while the size of the water body area in road images is not fixed, so directly using U-Net network for road image water body area detection will be difficult The precision and recall of the d...
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
Login to View More IPC IPC(8): G06K9/00G06N3/04G06N3/08
CPCG06N3/084G06V20/182G06N3/047G06N3/045
Inventor 王欢孟策
Owner NANJING UNIV OF SCI & TECH