Unlock instant, AI-driven research and patent intelligence for your innovation.

Pipeline disease image classification method based on multi-label convolutional neural network

A technology of convolutional neural network and classification method, which is applied in the field of computer digital image processing and deep learning algorithm based on convolutional neural network, which can solve the problems of different discrimination sensitivity, difficulty in obtaining high-quality images, and low classification accuracy. , to achieve a wide range of applications, rich types, and improved accuracy.

Active Publication Date: 2019-10-18
TIANHE COLLEGE GUANGDONG POLYTECHNIC NORMAL UNIV
View PDF6 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0017] However, the existing computer digital image processing technology needs to subjectively obtain the corresponding features of the disease image, and is too dependent on the quality of the captured image; for the dark, humid, and complex environment of the inner wall of the pipeline, it is difficult to obtain high-quality images;
[0018] Moreover, different disease types have different distinguishing sensitivities of image features, and what kind of features to choose as the classification basis requires a certain amount of experience in pipeline inspection, which limits the development of image detection technology;
[0019] In addition, due to its simple model and few classification types, the existing convolutional neural network cannot deal with complex pipeline disease classification problems, and the classification accuracy is low

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
  • Pipeline disease image classification method based on multi-label convolutional neural network
  • Pipeline disease image classification method based on multi-label convolutional neural network
  • Pipeline disease image classification method based on multi-label convolutional neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0046] A pipeline disease image classification method based on a multi-label convolutional neural network. The steps of the pipeline disease image classification method are as follows:

[0047] Step 1: Collect the pipe endoscope detection video, and extract the image frames in the pipe endoscope detection video;

[0048] Step 2: Calculate the timestamp feature of each image;

[0049] Step 3: Send a part of the image frames collected in step 1 to the multi-label convolutional neural network model for training, and obtain a multi-label convolutional neural network model that can correctly classify the types of pipeline diseases;

[0050] Step 4: Use the trained multi-label convolutional neural network model to detect the endoscopic image of the pipeline to be detected, and then the multi-label convolutional neural network model will output the one-hot code, and determine the existing pipeline disease type according to the one-hot code.

[0051] The multi-label convolutional neural network...

Embodiment 2

[0061] This embodiment combines the above-mentioned embodiments to further illustrate the content, so that those skilled in the art can more clearly understand the implementation of the present invention. The present invention adds multi-label classification to the existing Inception-ResNet-v2 network. Layer, realize the classification function of a variety of pipeline disease images, the present invention replaces the SoftMax classifier with a multi-label classification layer, so that Inception-ResNet-v2 has a multi-label classification function, thereby maximizing detection of the expected disease types ,Such as figure 2 Shown.

[0062] Among them, the random inactivation layer of the upper Inception-ResNet-v2 network structure will output a feature vector of 1792 dimensions, corresponding to figure 2 The X layer in. The present invention adds a one-dimensional feature to the back of this vector, such as figure 2 As shown by the light gray box in, this feature is the time...

Embodiment 3

[0069] This embodiment combines the above-mentioned embodiments to further illustrate the content, so that those skilled in the art can more clearly understand the implementation of the present invention. The present invention adds multi-label classification to the existing Inception-ResNet-v2 network. Layer, realize the classification function of a variety of pipeline disease images, the present invention replaces the SoftMax classifier with a multi-label classification layer, so that Inception-ResNet-v2 has a multi-label classification function, thereby maximizing detection of the expected disease types .

[0070] Such as figure 2 As shown, the X layer is the output vector of the random inactivation layer in the upper Inception-ResNet-v2 network structure shown.

[0071] The random inactivation layer of the original Inception-ResNet-v2 network structure only outputs the first 1792 dimensional information, and the present invention adds a dimension of time information on this ba...

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 pipeline disease image classification method based on a multi-label convolutional neural network. The pipeline disease image classification method comprises the following steps: 1, collecting a pipeline endoscopic detection video, and extracting an image frame in the video; 2, calculating a timestamp feature of each image; 3, sending part of the image frames collected inthe step 1 into a multi-label convolutional neural network model for training, and obtaining the multi-label convolutional neural network model capable of correctly classifying the pipeline disease types; and 4, detecting the endoscopic image of the pipeline to be detected by using the trained multi-label convolutional neural network model, then outputting a one-hot code by the multi-label convolutional neural network model, and determining the type of the existing pipeline disease according to the one-hot code. A multi-label classification layer is added on the basis of an existing Inception-ResNet-v2 network, and the classification function of various pipeline disease images is achieved.

Description

Technical field [0001] The invention relates to the field of computer digital image processing and deep learning algorithms based on convolutional neural networks, in particular to a pipeline disease image classification method based on multi-label convolutional neural networks. Background technique [0002] At present, there are two commonly used methods to obtain pipeline endoscopic images: one is the pipeline inspection robot technology based on CloseCircuit Television Inspection (CCTV), and the other is the pipeline periscope based on PipeQuickView Inspection (QV) technology. technology. [0003] The processing method after acquiring the image in the prior art is still mainly manual observation, and then classification and recognition. The following difficulties will arise in manual observation: [0004] The increase in the length of pipeline construction will result in the generation of huge endoscopic video and image data. Take the pipeline inspection robot as an example, its...

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): G06T7/00G06K9/62
CPCG06T7/0004G06T2207/10016G06T2207/20081G06T2207/20084G06T2207/30108G06F18/24147G06F18/241
Inventor 唐露新张宇维
Owner TIANHE COLLEGE GUANGDONG POLYTECHNIC NORMAL UNIV