Channel obstacle detection method, device, equipment and medium
By generating and expanding the data set in channel occupancy detection and training classification models, the accuracy problem when identifying channels are occupied by objects in traditional methods is solved, and higher recognition accuracy is achieved.
Patent Information
- Application Number
- CN202411919662.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-24
- Publication Date
- 2025-03-18
AI Technical Summary
When traditional channel occupancy detection methods face the problems of lack of negative samples, insufficient overall sample number and confusion in complex scene recognition, it is difficult to accurately identify the situation where the channel is occupied by objects in actual production scenarios.
By identifying channel ranges from different industrial scene pictures, combining target channel obstacles in the production workshop with channel ranges in these pictures, a channel occupancy data set is generated and the data set is expanded, including image geometric transformation, pixel filling, weighted averaging and enhancement. Then, the classification model is trained using the expanded data set to classify the images to be detected and obstacle detection.
It improves the accuracy of identifying channels occupied by objects in actual production scenarios, and solves the problems of lack of negative samples, insufficient overall sample number, and confusion in complex scene recognition.
Smart Images
Figure CN119649154A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of deep learning technology, and in particular to a channel obstacle detection method, device, equipment and medium. Background Art
[0002] Traditional occupancy detection methods usually use data from certain special scenes and areas for learning and training. Various special situations frequently occur in actual production scenarios, which seriously affect the accuracy of channel occupancy recognition. For example, sunlight of different intensities will greatly change the brightness and color of the environment, making recognition more difficult; occasional overexposure will cause information loss in some areas; moving shadows of objects or rapid shuttling of animals may be misjudged as channel occupation. These special situations pose huge challenges to traditional models during training and detection, and they are unable to accurately focus on the channel area, resulting in wrong judgments. At the same time, actual production shows an unbalanced distribution trend in terms of the number of samples, with deviations in the amount of occupied and non-occupied data, and tends to predict the category with the majority of numbers. This situation makes the traditional method prone to deviations in the training and prediction process, which greatly reduces the accuracy of channel occupancy recognition.
[0003] From the above, it can be seen that how to solve the problems of lack of negative samples, insufficient number of overall samples and confusion in complex scene recognition, and improve the accuracy of identifying channels occupied by objects in actual production scenarios are issues to be solved in this field. Summary of the invention
[0004] In view of this, the purpose of the present invention is to provide a channel obstacle detection method, device, equipment and medium, which can solve the problems of lack of negative samples, insufficient number of overall samples and confusion in complex scene recognition, and improve the accuracy of identifying channels occupied by objects in actual production scenarios. The specific scheme is as follows:
[0005] In a first aspect, the present application discloses a channel obstacle detection method, comprising:
[0006] Identify channel ranges from different industrial scene images, and combine target channel obstacles in the production workshop with the channel ranges in different industrial scene images to generate a channel occupancy data set; the target channel obstacle is an obstacle that appears in the channel with a historical frequency greater than a preset frequency threshold;
[0007] Expanding the channel occupancy data set to obtain an expanded channel occupancy data set; the data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement;
[0008] Using the expanded channel occupancy data set to train a preset classification model to obtain the trained classification model;
[0009] Based on the trained classification model, the image of the channel to be detected is classified and obstacles are detected to complete the channel obstacle detection of the channel to be detected.
[0010] Optionally, identifying channel ranges from different industrial scene images, combining target channel obstacles in the production workshop with the channel ranges in different industrial scene images to generate a channel occupancy data set includes:
[0011] Use image processing software and segmentation algorithms to cut out target channel obstacles from channel-occupying objects in the production workshop to generate an obstacle set and screen out different industrial scene images;
[0012] The channel range is identified from different industrial scene pictures, and the target channel obstacles in the obstacle set are randomly mapped to the channel range in the different industrial scene pictures to generate a channel occupancy data set.
[0013] Optionally, identifying a channel range from different industrial scene pictures, and randomly mapping a target channel obstacle in the obstacle set to the channel range in the different industrial scene pictures includes:
[0014] Using edge detection algorithms to identify channel ranges from different industrial scene images and determine the coordinate range of the channel range;
[0015] A random number is set, and the target channel obstacles corresponding to the random number in the obstacle set are randomly mapped into the coordinate range.
[0016] Optionally, the step of expanding the channel occupancy data set to obtain an expanded channel occupancy data set includes:
[0017] Performing image geometric transformation on the channel occupancy data set, and filling pixels of the transformed image using a bilinear interpolation method to obtain a filled image;
[0018] The filled image is scaled, and Gaussian blur and sharpening algorithms are used for weighted averaging and convolution superposition to obtain a superimposed image;
[0019] The brightness of the superimposed image is calculated using an unsupervised backlight enhancement algorithm, and the pixel points are enhanced using an edge detection operator to obtain the expanded channel occupancy data set.
[0020] Optionally, before performing obstacle detection and classification on the image of the channel to be detected based on the trained classification model, the method further includes:
[0021] A classification model is built based on a sample classifier, multiple convolutional layers, a self-attention mechanism layer, a feature layer, a layer normalization layer, a decision function, and a classification detection head module.
[0022] Optionally, the classifying and obstacle detection of the image of the channel to be detected based on the trained classification model to complete the channel obstacle detection of the channel to be detected includes:
[0023] The Edge Next algorithm is used and based on the self-attention mechanism layer, feature layer, layer normalization layer, single target domain adaptation module, neural network module, and classification layer in the trained classification model to perform calculations, deep separable convolution, normalization, dynamic multi-scale feature fusion and global information encoding, feature fusion, and classification on the image of the channel to be detected to complete the channel obstacle detection of the channel to be detected.
[0024] In a second aspect, the present application discloses a channel obstacle detection device, comprising:
[0025] A data set generation module, used to identify channel ranges from different industrial scene images, and combine target channel obstacles in the production workshop with the channel ranges in different industrial scene images to generate a channel occupancy data set; the target channel obstacle is an obstacle whose historical frequency of appearance in the channel is greater than a preset frequency threshold;
[0026] A data set expansion module, used for expanding the channel occupancy data set to obtain the expanded channel occupancy data set; the data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement;
[0027] A model training module, used to train a preset classification model using the expanded channel occupancy data set to obtain the trained classification model;
[0028] The obstacle detection module is used to classify the image of the channel to be detected and detect obstacles based on the trained classification model to complete the channel obstacle detection of the channel to be detected.
[0029] Optionally, the data set generation module includes:
[0030] The obstacle set generation module is used to cut out the target channel obstacles from the channel-occupying objects in the production workshop using image processing software and segmentation algorithms to generate an obstacle set and screen out different industrial scene images;
[0031] The channel occupancy data set generation module is used to identify the channel range from different industrial scene pictures, and randomly map the target channel obstacles in the obstacle set to the channel range in different industrial scene pictures to generate a channel occupancy data set.
[0032] In a third aspect, the present application discloses an electronic device, comprising:
[0033] Memory, used to store computer programs;
[0034] The processor is used to execute the computer program to implement the aforementioned channel obstacle detection method.
[0035] In a fourth aspect, the present application discloses a computer storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the steps of the channel obstacle detection method disclosed above are implemented.
[0036] It can be seen that the present application provides a channel obstacle detection method, including identifying channel ranges from different industrial scene pictures, combining target channel obstacles in a production workshop with the channel ranges in different industrial scene pictures to generate a channel occupancy data set; the target channel obstacle is an obstacle whose historical frequency of appearance in the channel is greater than a preset frequency threshold; the channel occupancy data set is expanded to obtain an expanded channel occupancy data set; the data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement; a preset classification model is trained using the expanded channel occupancy data set to obtain the trained classification model; based on the trained classification model, the image of the channel to be detected is classified and obstacle detection is performed to complete the channel obstacle detection of the channel to be detected. In view of the complex and changeable environmental conditions in actual application scenarios, even the same workspace scene will be affected by changes in the position of the sun, overexposure of reflections, and occlusion by walking people. The application identifies channel ranges from different industrial scene pictures, combines target channel obstacles in the production workshop with the channel ranges in different industrial scene pictures to generate a channel occupancy data set, and expands the channel occupancy data set to address the sample imbalance problem caused by the small number of data on channel occupancy in actual production workshops; the data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement. The expanded channel occupancy data set is used to train a classification model, and the images of the channel to be detected are classified and obstacle detection is performed based on the trained classification model to complete the channel obstacle detection of the channel to be detected. A channel occupancy intelligent recognition solution based on deep learning is proposed to optimize the channel management of production workshops and work platforms, which can solve the problems of lack of negative samples, insufficient overall sample quantity, and confusion in complex scene recognition, and improve the accuracy of identifying channels occupied by objects in actual production scenarios. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0038] Figure 1 A flow chart of a channel obstacle detection method disclosed in this application;
[0039] Figure 2 A flow chart for generating a channel occupancy data set disclosed in this application;
[0040] Figure 3This is an overall framework diagram of a channel obstacle detection disclosed in this application;
[0041] Figure 4 This is a schematic diagram of the structure of a channel obstacle detection device disclosed in this application;
[0042] Figure 5 A structural diagram of an electronic device provided for this application. DETAILED DESCRIPTION
[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0044] Traditional occupancy detection methods usually learn and train data from certain special scenes and areas. Various special situations frequently occur in actual production scenes, which seriously affect the accuracy of channel occupancy recognition. For example, sunlight of different intensities will greatly change the brightness and color of the environment, making recognition more difficult; occasional overexposure will cause information loss in some areas; moving shadows of objects or rapid shuttles of animals may be misjudged as channel occupation. These special situations make traditional models face huge challenges in the training and detection process, and they cannot accurately focus on the channel area, resulting in wrong judgments. At the same time, actual production shows an unbalanced distribution trend in terms of the number of samples, and there is a deviation in the amount of occupied and non-occupied data, which tends to predict the category with the dominant number. This situation makes the traditional method prone to deviations in the training and prediction process, which greatly reduces the accuracy of channel occupancy recognition. As can be seen from the above, how to solve the problems of lack of negative samples, insufficient number of overall samples, and confusion in complex scene recognition, and improve the accuracy of identifying channels occupied by objects in actual production scenes is a problem to be solved in this field.
[0045] See also Figure 1 As shown, an embodiment of the present invention discloses a channel obstacle detection method, which may specifically include:
[0046] Step S11: Identify the channel range from different industrial scene pictures, and combine the target channel obstacle in the production workshop with the channel range in different industrial scene pictures to generate a channel occupancy data set; the target channel obstacle is an obstacle whose historical frequency of appearance in the channel is greater than a preset frequency threshold.
[0047] In this embodiment, image processing software and segmentation algorithms are used to cut out target channel obstacles from channel-occupying objects in a production workshop to generate an obstacle set, and different industrial scene pictures are screened out; channel ranges are identified from different industrial scene pictures, and target channel obstacles in the obstacle set are randomly mapped to the channel ranges in different industrial scene pictures to generate a channel occupancy data set.
[0048] Among them, the specific process of randomly mapping the target channel obstacles is: using the edge detection algorithm to identify the channel range from different industrial scene pictures, and determine the coordinate range of the channel range; setting a random number, and randomly mapping the target channel obstacles corresponding to the random number in the obstacle set to the coordinate range.
[0049] In actual production workshops, the data of occupied channels is far less than that of unobstructed channels. Serious sample imbalance problems will occur during model training, resulting in high accuracy of the trained model for unobstructed channels and poor recognition performance for occupied channels. Artificial data generation is required to reduce the difference in quantity. The generation process of channel occupancy dataset is as follows: Figure 2 As shown in the figure, firstly, Adobe Photoshop and segmentation algorithm are used to cut out the common occupied objects in the channel (i.e., the target channel obstacles in the production workshop) to create an obstacle set. Next, we select a variety of typical scenes from high-quality positive samples (i.e., different industrial scene pictures) as the original set . Start from the original set P and traverse it. For each graph After completing a random mapping, the algorithm sets a random number to extract 1-2 objects from the obstacle set O. (i=1,2), calculate the effective area based on the length and width C and W of the object (i.e. the coordinate range corresponding to the channel range) to ensure the integrity of the map, in the valid area A random position is generated inside the image to complete the mapping. This method can generate a large number of negative samples, and then through simple manual screening, channel occupancy data sets in various scenarios can be simulated and produced.
[0050] Step S12: expanding the channel occupancy data set to obtain an expanded channel occupancy data set; the data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement.
[0051] In this embodiment, the process of data set expansion is as follows: performing image geometric transformation on the channel occupancy data set, and using bilinear interpolation to fill pixels of the transformed image to obtain a filled image; scaling the filled image, and using Gaussian blur and sharpening algorithms to perform weighted averaging and convolution superposition to obtain a superimposed image; using an unsupervised backlight enhancement algorithm to calculate the brightness of the superimposed image, and using an edge detection operator to perform pixel enhancement to obtain the expanded channel occupancy data set.
[0052] This application mainly realizes data expansion through geometric transformation, pixel operation and unsupervised backlight enhancement algorithm. In geometric transformation, the rotation operation first calculates the rotation matrix to process the image, and then uses bilinear interpolation to fill the pixels outside the image boundary to ensure the integrity and rationality of the rotated image. The scaling operation uses a scale factor of 0.8 to 1.2 to randomly process the image. This operation helps the subsequent classification model learn the characteristic patterns of objects at different scales. In terms of pixel operation, Gaussian blur and sharpening algorithms are used. Among them, Gaussian blur uses a 3×3 Gaussian kernel and uses a Gaussian function to perform weighted averaging on each pixel in the image and its neighboring pixels, so that the high-frequency detail information (such as edges, noise, etc.) in the image is weakened, thereby achieving the effect of blurring the image. The sharpening algorithm uses the Laplace sharpening method, which constructs a Laplace operator and performs a convolution operation with the image, and then superimposes the convolution result with the original image. After multiple experimental parameter adjustments, the optimal operator coefficient is selected to ensure the appropriate sharpening strength. The last expansion method is the unsupervised backlight enhancement algorithm, which calculates the brightness of the input RGB (Red Green Blue) samples. This deep learning algorithm adaptively adjusts the graphic histogram according to the regional color domain, brightness domain, etc., completes a certain degree of denoising and color domain adjustment through unsupervised processing, and realizes pixel-level enhancement of the edge part through the Sobel operator in the edge detection operator. Through the above-mentioned multiple methods, the number of samples is enriched and the data set is expanded, which improves the perception ability of the subsequent classification detection model to a certain extent.
[0053] Step S13: using the expanded channel occupancy data set to train the preset classification model to obtain the trained classification model.
[0054] In this embodiment, a classification model is constructed based on a sample classifier, multiple convolutional layers, a self-attention mechanism layer, a feature layer, a layer normalization layer, a decision function, and a classification detection head module, and the preset classification model is trained using the expanded channel occupancy data set.
[0055] In this embodiment, the expanded channel occupancy data set is fed into the classification model to learn and train a sample classifier. Through a series of convolutional layers, channel features are deeply mined, and feature data is distinguished through decision functions. Finally, the samples are classified into four categories: unoccupied, occupied, exposed, and occluded by people. The model can choose a variety of image classification algorithms, such as YOLOv series, Edge Next, EfficientNet, etc.
[0056] Step S14: classifying the image of the channel to be detected and detecting obstacles based on the trained classification model to complete the channel obstacle detection of the channel to be detected.
[0057] In this embodiment, the Edge Next algorithm is used and based on the self-attention mechanism layer, feature layer, layer normalization layer, single target domain adaptation module, neural network module, and classification layer in the trained classification model to perform calculations, deep separable convolution, normalization, dynamic multi-scale feature fusion and global information encoding, feature fusion, and classification on the image of the channel to be detected to complete the channel obstacle detection of the channel to be detected.
[0058] This application uses the lightweight Edge Next algorithm as an example of the key classification algorithm. It uses dynamic convolution kernel size to improve the model's receptive field for channel images while reducing the amount of calculation. It also applies Self-attention to the channel dimension of RGB images for the first time, obtains linear complexity, and further reduces the amount of calculation. The training set P is sent to the network, and the image is cut into feature maps of different resolutions through 4 stages, effectively expanding the range of the receptive field. After the feature layer passes through the Depthwise Separable Convolution, the LN (LayerNormalization) layer is used for normalization, and then a 1×1 projection convolution is used to achieve feature enhancement and dimensional transformation. The feature map is then sent to the STDA (single target domain adaptation) module to realize dynamic multi-scale feature fusion and global information encoding, and the linear level feature fusion is realized by transposing the self-attention transformer (neural network) module. Finally, the high-dimensional abstract features are sent to the classification layer to obtain the category score. After 400 iterations of training, the recognition accuracy of occupied samples reached 89.64%, the inference speed was about 12ms, and the memory occupied was less than 1.5G.
[0059] The channel obstacle detection method proposed in this application consists of three parts: channel occupancy data set, data set expansion, and classification and recognition. The overall framework is as follows: Figure 3As shown in the figure, the first two parts belong to the data processing part. First, the corresponding channel occupancy data set is generated based on the original unoccupied image, which can increase enough occupancy samples for the actual production workshop. Then, the effective data samples are expanded by using backlight enhancement, scaling, Gaussian blur, color gamut adjustment and other methods to obtain the expanded channel occupancy data set. The last part is the obstacle detection part. First, the video frames of the channel to be detected that are difficult to judge under exposure or darkness are filtered, and finally sent to the trained classification model to classify and determine whether it is occupied through the neural network model.
[0060] This application proposes a channel occupancy identification method based on a deep learning classification model. It starts with the data set, improves the model classification logic and formulates an effective plan, and studies a new route for the occupancy identification method of the production workshop, which is effectively applied to the channel occupancy detection task in industrial scenarios such as production workshops; this patent provides an effective idea for generating negative samples and data expansion; this application has strong generalization capabilities under exposure, human occlusion, and multiple scenes, and can solve the problems of lack of negative samples, insufficient overall sample quantity, and confusion in complex scene recognition, and ultimately successfully improves the accuracy of identifying object occupancy in actual production scenarios.
[0061] In this embodiment, the channel range is identified from different industrial scene pictures, and the target channel obstacle in the production workshop is combined with the channel range in different industrial scene pictures to generate a channel occupancy data set; the target channel obstacle is an obstacle whose historical frequency of appearance in the channel is greater than a preset frequency threshold; the channel occupancy data set is expanded to obtain the expanded channel occupancy data set; the data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement; the preset classification model is trained using the expanded channel occupancy data set to obtain the trained classification model; based on the trained classification model, the image of the channel to be detected is classified and obstacle detection is performed to complete the channel obstacle detection of the channel to be detected. In view of the complex and changeable environmental conditions in actual application scenarios, even the same workspace scene will be affected by changes in the position of the sun, overexposure of reflections, and occlusion by walking people. The application identifies channel ranges from different industrial scene pictures, combines target channel obstacles in the production workshop with the channel ranges in different industrial scene pictures to generate a channel occupancy data set, and expands the channel occupancy data set to address the sample imbalance problem caused by the small number of data on channel occupancy in actual production workshops; the data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement. The expanded channel occupancy data set is used to train a classification model, and the images of the channel to be detected are classified and obstacle detection is performed based on the trained classification model to complete the channel obstacle detection of the channel to be detected. A channel occupancy intelligent recognition solution based on deep learning is proposed to optimize the channel management of production workshops and work platforms, which can solve the problems of lack of negative samples, insufficient overall sample quantity, and confusion in complex scene recognition, and improve the accuracy of identifying channels occupied by objects in actual production scenarios.
[0062] See also Figure 4 As shown, an embodiment of the present invention discloses a channel obstacle detection device, which may specifically include:
[0063] The data set generation module 11 is used to identify the channel range from different industrial scene pictures, and combine the target channel obstacle in the production workshop with the channel range in different industrial scene pictures to generate a channel occupancy data set; the target channel obstacle is an obstacle whose historical frequency of appearance in the channel is greater than a preset frequency threshold;
[0064] The data set expansion module 12 is used to expand the channel occupancy data set to obtain the expanded channel occupancy data set; the data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement;
[0065] A model training module 13, used to train a preset classification model using the expanded channel occupancy data set to obtain the trained classification model;
[0066] The obstacle detection module 14 is used to classify the image of the channel to be detected and detect obstacles based on the trained classification model to complete the channel obstacle detection of the channel to be detected.
[0067] In some specific embodiments, the data set generating module 11 may specifically include:
[0068] The obstacle set generation module is used to cut out the target channel obstacles from the channel-occupying objects in the production workshop using image processing software and segmentation algorithms to generate an obstacle set and screen out different industrial scene images;
[0069] The channel occupancy data set generation module is used to identify the channel range from different industrial scene pictures, and randomly map the target channel obstacles in the obstacle set to the channel range in different industrial scene pictures to generate a channel occupancy data set.
[0070] In some specific embodiments, the data set generating module 11 may specifically include:
[0071] A coordinate range determination module, used to identify channel ranges from different industrial scene images using an edge detection algorithm, and determine the coordinate range of the channel range;
[0072] The random mapping module is used to set a random number and randomly map the target channel obstacles corresponding to the random number in the obstacle set to the coordinate range.
[0073] In some specific embodiments, the data set expansion module 12 may specifically include:
[0074] An image geometric transformation and pixel filling module, used to perform image geometric transformation on the channel occupancy data set, and use a bilinear interpolation method to fill pixels on the transformed image to obtain a filled image;
[0075] The superposition module is used to scale the filled image and use Gaussian blur and sharpening algorithms to perform weighted averaging and convolution superposition to obtain a superimposed image;
[0076] The brightness calculation and pixel enhancement module is used to perform brightness calculation on the superimposed image using an unsupervised backlight enhancement algorithm and to perform pixel enhancement using an edge detection operator to obtain the expanded channel occupancy data set.
[0077] In some specific embodiments, the obstacle detection module 14 may specifically include:
[0078] The classification model building module is used to build a classification model based on a sample classifier, multiple convolutional layers, a self-attention mechanism layer, a feature layer, a layer normalization layer, a decision function, and a classification detection head module.
[0079] In some specific embodiments, the obstacle detection module 14 may specifically include:
[0080] The obstacle category detection module is used to use the Edge Next algorithm and based on the self-attention mechanism layer, feature layer, layer normalization layer, single target domain adaptation module, neural network module, and classification layer in the trained classification model to perform calculations, deep separable convolution, normalization, dynamic multi-scale feature fusion and global information encoding, feature fusion, and classification on the image of the channel to be detected, so as to complete the channel obstacle detection of the channel to be detected.
[0081] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the channel obstacle detection method performed by the electronic device disclosed in any of the aforementioned embodiments.
[0082] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.
[0083] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon include an operating system 221, a computer program 222 and data 223, etc. The storage method can be temporary storage or permanent storage.
[0084] Among them, the operating system 221 is used to manage and control the hardware devices and computer programs 222 on the electronic device 20 to realize the operation and processing of the data 223 in the memory 22 by the processor 21, which can be Windows, Unix, Linux, etc. In addition to including a computer program that can be used to complete the channel obstacle detection method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks. In addition to data transmitted from an external device received by the channel obstacle detection device, the data 223 can also include data collected by its own input and output interface 25, etc.
[0085] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
[0086] Furthermore, an embodiment of the present application also discloses a computer-readable storage medium, in which a computer program is stored. When the computer program is loaded and executed by a processor, the steps of the channel obstacle detection method disclosed in any of the aforementioned embodiments are implemented.
[0087] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
[0088] The above is a detailed introduction to a channel obstacle detection method, device, equipment and storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.
Claims
1. A channel obstacle detection method, characterized in that: include: Identify channel ranges from different industrial scene images, and combine target channel obstacles in the production workshop with the channel ranges in different industrial scene images to generate a channel occupancy data set; the target channel obstacle is an obstacle that appears in the channel with a historical frequency greater than a preset frequency threshold; Expanding the channel occupancy data set to obtain an expanded channel occupancy data set; The data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement; Using the expanded channel occupancy data set to train a preset classification model to obtain the trained classification model; Based on the trained classification model, the image of the channel to be detected is classified and obstacles are detected to complete the channel obstacle detection of the channel to be detected.
2. The channel obstacle detection method according to claim 1, characterized in that: The identifying channel ranges from different industrial scene images, combining the target channel obstacles in the production workshop with the channel ranges in different industrial scene images to generate a channel occupancy data set, includes: Use image processing software and segmentation algorithms to cut out target channel obstacles from channel-occupying objects in the production workshop to generate an obstacle set and screen out different industrial scene images; The channel range is identified from different industrial scene pictures, and the target channel obstacles in the obstacle set are randomly mapped to the channel range in the different industrial scene pictures to generate a channel occupancy data set.
3. The channel obstacle detection method according to claim 2, characterized in that: The identifying channel ranges from different industrial scene pictures and randomly mapping target channel obstacles in the obstacle set to the channel ranges in different industrial scene pictures includes: Using edge detection algorithms to identify channel ranges from different industrial scene images and determine the coordinate range of the channel range; A random number is set, and the target channel obstacles corresponding to the random number in the obstacle set are randomly mapped into the coordinate range.
4. The channel obstacle detection method according to claim 1, characterized in that: The step of expanding the channel occupancy data set to obtain the expanded channel occupancy data set includes: Performing image geometric transformation on the channel occupancy data set, and filling pixels of the transformed image using a bilinear interpolation method to obtain a filled image; The filled image is scaled, and Gaussian blur and sharpening algorithms are used for weighted averaging and convolution superposition to obtain a superimposed image; The brightness of the superimposed image is calculated using an unsupervised backlight enhancement algorithm, and the pixel points are enhanced using an edge detection operator to obtain the expanded channel occupancy data set.
5. The channel obstacle detection method according to claim 1, characterized in that: Before performing obstacle detection and classification on the image of the detection channel based on the trained classification model, the method further includes: A classification model is built based on a sample classifier, multiple convolutional layers, a self-attention mechanism layer, a feature layer, a layer normalization layer, a decision function, and a classification detection head module.
6. The channel obstacle detection method according to any one of claims 1 to 5, characterized in that: The image of the channel to be detected is classified and the obstacle detection is performed based on the trained classification model to complete the channel obstacle detection of the channel to be detected, including: The Edge Next algorithm is used and based on the self-attention mechanism layer, feature layer, layer normalization layer, single target domain adaptation module, neural network module, and classification layer in the trained classification model to perform calculations, deep separable convolution, normalization, dynamic multi-scale feature fusion and global information encoding, feature fusion, and classification on the image of the channel to be detected to complete the channel obstacle detection of the channel to be detected.
7. A channel obstacle detection device, characterized in that: include: A data set generation module, used to identify channel ranges from different industrial scene images, and combine target channel obstacles in the production workshop with the channel ranges in different industrial scene images to generate a channel occupancy data set; the target channel obstacle is an obstacle whose historical frequency of appearance in the channel is greater than a preset frequency threshold; A data set expansion module, used for expanding the channel occupancy data set to obtain the expanded channel occupancy data set; the data set expansion includes image geometric transformation, pixel filling, pixel weighted averaging, and pixel enhancement; A model training module, used to train a preset classification model using the expanded channel occupancy data set to obtain the trained classification model; The obstacle detection module is used to classify the image of the channel to be detected and detect obstacles based on the trained classification model to complete the channel obstacle detection of the channel to be detected.
8. The passage obstacle detection device according to claim 7, characterized in that: The data set generation module comprises: The obstacle set generation module is used to cut out the target channel obstacles from the channel-occupying objects in the production workshop using image processing software and segmentation algorithms to generate an obstacle set and screen out different industrial scene images; The channel occupancy data set generation module is used to identify the channel range from different industrial scene pictures, and randomly map the target channel obstacles in the obstacle set to the channel range in different industrial scene pictures to generate a channel occupancy data set.
9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the channel obstacle detection method as described in any one of claims 1 to 6.
10. A computer-readable storage medium, characterized in that: Used to store a computer program; wherein, when the computer program is executed by a processor, the channel obstacle detection method according to any one of claims 1 to 6 is implemented.