Deepwater net damage identification method, device, equipment and storage medium
By employing underwater image enhancement and unsupervised anomaly detection methods, the problems of low local image contrast and difficulty in threshold acquisition in underwater netting damage detection are solved, achieving highly accurate and applicable netting damage identification.
Patent Information
- Application Number
- CN202111497285.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-09
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2041-12-09
AI Technical Summary
In existing technologies, underwater netting damage detection suffers from problems such as low local image contrast leading to false detections and difficulty in obtaining thresholds, affecting detection accuracy and applicability.
A method based on underwater image enhancement and unsupervised anomaly detection is adopted, including grayscale processing, image blending enhancement, local adaptive threshold binarization and connected component segmentation, combined with the unsupervised anomaly detection algorithm HBOS to extract the location of the netting damage.
It improves the accuracy and applicability of deep-sea netting damage detection, reduces the probability of false detection, and simplifies the detection process.
Smart Images

Figure CN114419533B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of fishery farming, and in particular to a method, device, equipment and computer-readable storage medium for identifying damage to deep-water nets. Background Art
[0002] With the accelerated transformation of aquaculture from offshore aquaculture to deep-sea aquaculture, from cage aquaculture to large-scale intelligent equipment aquaculture, and from traditional artificial aquaculture to automated industrial aquaculture, unmanned daily monitoring and maintenance of large-scale aquaculture equipment must be considered. As an important component of deep-sea aquaculture cages, one of the main problems in daily maintenance is the determination of the location and degree of damage to the underwater nets. Traditional manual inspections have efficiency and safety issues, so unmanned detection technology for underwater nets is in urgent need in the field of deep-sea aquaculture. At present, vision-based underwater net detection technology has achieved certain development, but there are still prominent problems in actual application. For example:
[0003] (1) During the image preprocessing process, due to factors such as uneven underwater illumination, the local contrast of the image is low, and the processing results often lose image details, resulting in false detection of damaged points on the net;
[0004] (2) In the process of identifying damaged points of nets based on their characteristic values, it is generally necessary to set a threshold to extract abnormal values. However, the threshold should be obtained based on the characteristics of the nets and breeding experience, and there is no clear standard, which limits the applicability of the damage detection algorithm.
[0005] In view of the problems existing in the above-mentioned image preprocessing and damage anomaly detection, the present invention proposes a deep-water net damage identification method based on underwater image enhancement and unsupervised anomaly detection. Summary of the invention
[0006] The main purpose of the present invention is to provide a method for identifying damaged deep-water nets, aiming to solve the technical problem of easy misdetection of damaged nets in the prior art.
[0007] To achieve the above object, the present invention provides a method for identifying damaged deep-water nets, the method comprising:
[0008] The collected original image of deep-water net is gray-processed to obtain a single-channel gray-scale deep-water net image;
[0009] Performing image hybrid enhancement processing on the single-channel grayscale deep-water net image to obtain a hybrid enhanced deep-water net image;
[0010] Performing local adaptive threshold binarization processing on the hybrid enhanced deep-water net image to obtain a binarized deep-water net image;
[0011] Performing connected domain segmentation processing on the binary deep-water net image to obtain a mesh label sequence and a mesh area sequence;
[0012] An unsupervised anomaly detection process is performed on the mesh area sequence to extract anomalies, and a mesh damage recognition result is determined according to mesh labels corresponding to the anomalies.
[0013] Optionally, the step of performing image mixing enhancement processing on the single-channel grayscale deep-water net image comprises:
[0014] Performing a guided filtering operation on the single-channel grayscale deep-water net image using a window image of a preset size;
[0015] Using limited contrast adaptive histogram equalization to perform a local contrast enhancement processing operation on the window image after the guided filtering operation;
[0016] The window image after the local contrast enhancement processing operation is filtered again with the window image of a preset size.
[0017] Optionally, the step of obtaining the hybrid enhanced deep-water net image comprises:
[0018] Slide the window image, and after each slide, perform the step of performing image mixed enhancement processing on the single-channel grayscale deep-water net image, until the single-channel grayscale deep-water net image is traversed to obtain the mixed enhancement processed deep-water net image.
[0019] Optionally, the step of performing local adaptive threshold binarization processing on the hybrid enhanced deep-water net image comprises:
[0020] The hybrid enhanced deep-water net image is binarized and inverted using a local adaptive threshold binarization algorithm, and the hybrid enhanced deep-water net image is simplified into a binary image, in which the background is the net wire and the target is the mesh.
[0021] Optionally, the step of performing connected domain segmentation processing on the binary deep-water net image comprises:
[0022] Using an eight-neighborhood indexing method, each mesh is identified as a connected domain, and each connected domain is marked;
[0023] The number of pixels in each mesh is counted as the area of the corresponding mesh, and the meshes at the edge of the binary deep-water net image and the meshes whose area is smaller than the preset background area are set as the background.
[0024] Optionally, the step of determining the net damage identification result according to the mesh label corresponding to the abnormal value includes:
[0025] Determine the damaged position of the net according to the mesh label corresponding to the abnormal value, and calculate the average value of the mesh area of the connected domain around the damaged position of the net;
[0026] The ratio of the mesh area at the damaged position of the net to the average mesh area is taken as the net damage detection result.
[0027] Optionally, the step of calculating the mean value of the mesh area of the connected domain around the damaged position of the net includes:
[0028] The adjacent positions of the damaged net clothing position are selected according to the damaged net clothing position, and the average value of the mesh areas of the adjacent positions is used as the average value of the mesh areas of the connected domain surrounding the damaged net clothing position.
[0029] In addition, in order to achieve the above-mentioned purpose, the present invention also provides a deep-water net damage identification device, the deep-water net damage identification device comprising:
[0030] A grayscale processing module is used to perform grayscale processing on the collected original image of deep-water net to obtain a single-channel grayscale deep-water net image;
[0031] An image hybrid enhancement processing module is used to perform image hybrid enhancement processing on the single-channel grayscale deep-water net image to obtain a hybrid enhanced deep-water net image;
[0032] A local adaptive threshold binarization processing module is used to perform local adaptive threshold binarization processing on the hybrid enhanced deep-water net image to obtain a binarized deep-water net image;
[0033] A connected domain segmentation processing module is used to perform connected domain segmentation processing on the binary deep-water net image to obtain a mesh label sequence and a mesh area sequence;
[0034] The unsupervised anomaly detection processing module is used to perform unsupervised anomaly detection processing on the mesh area sequence, extract abnormal values, and determine the mesh damage recognition result according to the mesh labels corresponding to the abnormal values.
[0035] In addition, to achieve the above-mentioned purpose, the present invention also provides a deep-water net damage identification device, which includes: a memory, a processor, and a deep-water net damage identification program stored in the memory and executable on the processor, and the deep-water net damage identification program implements the steps of the deep-water net damage identification method as described above when executed by the processor.
[0036] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which a deep-water net damage identification program is stored, and when the deep-water net damage identification program is executed by a processor, the steps of the deep-water net damage identification method as described above are implemented.
[0037] The present invention proposes a method, device and computer-readable storage medium for identifying deep-water net damage. In one embodiment of the deep-water net damage identification method, in order to overcome the shortcomings of low local contrast of underwater net images and difficulty in extracting outliers in the net damage detection process, a net damage identification method based on underwater image enhancement and unsupervised anomaly detection is adopted to improve the accuracy and applicability of deep-water cage net damage detection. The underwater image enhancement technology is a "sandwich" type underwater net image hybrid enhancement technology, that is, guided filtering, contrast-limited adaptive histogram equalization, and guided filtering are performed on the image in the form of window sliding, and the local adaptive threshold binarization algorithm is used to perform binarization operations on the image, which effectively overcomes the problem of low local contrast of underwater images and reduces the probability of false detection of damaged meshes; at the same time, an unsupervised anomaly detection algorithm is introduced to simplify the net damage detection process and improve applicability. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a schematic diagram of the structure of the operating device of the hardware operating environment involved in the embodiment of the present invention;
[0039] Figure 2 It is a schematic diagram of the process of an embodiment of a method for identifying damage to deep-water nets of the present invention;
[0040] Figure 3 This is a diagram showing the filtering effect of using different filters on an underwater net image in an embodiment of the present invention;
[0041] Figure 4 is a grayscale histogram and a cumulative histogram of an original image and an image processed by CLAHE in an embodiment of the present invention;
[0042] Figure 5 The original image and the binarized image after image enhancement in the embodiment of the present invention;
[0043] Figure 6 This is a schematic diagram of the connected domain segmentation result of an underwater net image according to an embodiment of the present invention;
[0044] Figure 7 is a mesh area distribution histogram in an embodiment of the present invention;
[0045] Figure 8 This is the detection result of the damage of the net in the embodiment of the present invention.
[0046] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION
[0047] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0048] Reference Figure 1 , Figure 1 It is a schematic diagram of the structure of the operating equipment of the hardware operating environment involved in the embodiment of the present invention.
[0049] like Figure 1 As shown, the operating device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Among them, the communication bus 1002 is used to realize the connection and communication between these components. The user interface 1003 may include a display screen (Display), an input unit such as a keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface and a wireless interface. The network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a wireless fidelity (WIreless-FIdelity, WI-FI) interface). The memory 1005 may be a high-speed random access memory (Random Access Memory, RAM) memory, or a stable non-volatile memory (Non-Volatile Memory, NVM), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.
[0050] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation on the operating device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.
[0051] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a data storage module, a network communication module, a user interface module and a deep-water net damage identification program.
[0052] exist Figure 1In the running device shown, the network interface 1004 is mainly used for data communication with other devices; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the running device of the present invention can be set in the running device, and the running device calls the deep-water net damage identification program stored in the memory 1005 through the processor 1001, and performs the following operations:
[0053] The collected original image of deep-water net is gray-processed to obtain a single-channel gray-scale deep-water net image;
[0054] Performing image hybrid enhancement processing on the single-channel grayscale deep-water net image to obtain a hybrid enhanced deep-water net image;
[0055] Performing local adaptive threshold binarization processing on the hybrid enhanced deep-water net image to obtain a binarized deep-water net image;
[0056] Performing connected domain segmentation processing on the binary deep-water net image to obtain a mesh label sequence and a mesh area sequence;
[0057] An unsupervised anomaly detection process is performed on the mesh area sequence to extract anomalies, and a mesh damage recognition result is determined according to mesh labels corresponding to the anomalies.
[0058] Furthermore, the processor 1001 may call the deep-water net damage identification program stored in the memory 1005, and further perform the following operations:
[0059] The step of performing image mixing enhancement processing on the single-channel grayscale deep-water net image comprises:
[0060] Performing a guided filtering operation on the single-channel grayscale deep-water net image using a window image of a preset size;
[0061] Using limited contrast adaptive histogram equalization to perform a local contrast enhancement processing operation on the window image after the guided filtering operation;
[0062] The window image after the local contrast enhancement processing operation is filtered again with the window image of a preset size.
[0063] Furthermore, the processor 1001 may call the deep-water net damage identification program stored in the memory 1005, and further perform the following operations:
[0064] The step of obtaining the hybrid enhanced deep-water net image comprises:
[0065] Slide the window image, and after each slide, perform the step of performing image mixed enhancement processing on the single-channel grayscale deep-water net image, until the single-channel grayscale deep-water net image is traversed to obtain the mixed enhancement processed deep-water net image.
[0066] Furthermore, the processor 1001 may call the deep-water net damage identification program stored in the memory 1005, and further perform the following operations:
[0067] The step of performing local adaptive threshold binarization processing on the hybrid enhanced deep-water net image comprises:
[0068] The hybrid enhanced deep-water net image is binarized and inverted using a local adaptive threshold binarization algorithm, and the hybrid enhanced deep-water net image is simplified into a binary image, in which the background is the net wire and the target is the mesh.
[0069] Furthermore, the processor 1001 may call the deep-water net damage identification program stored in the memory 1005, and further perform the following operations:
[0070] The step of performing connected domain segmentation processing on the binary deep-water net image comprises:
[0071] Using an eight-neighborhood indexing method, each mesh is identified as a connected domain, and each connected domain is marked;
[0072] The number of pixels in each mesh is counted as the area of the corresponding mesh, and the meshes at the edge of the binary deep-water net image and the meshes whose area is smaller than the preset background area are set as the background.
[0073] Furthermore, the processor 1001 may call the deep-water net damage identification program stored in the memory 1005, and further perform the following operations:
[0074] The step of determining the net damage identification result according to the mesh label corresponding to the abnormal value comprises:
[0075] Determine the damaged position of the net according to the mesh label corresponding to the abnormal value, and calculate the average value of the mesh area of the connected domain around the damaged position of the net;
[0076] The ratio of the mesh area at the damaged position of the net to the average mesh area is taken as the net damage detection result.
[0077] Furthermore, the processor 1001 may call the deep-water net damage identification program stored in the memory 1005, and further perform the following operations:
[0078] The step of calculating the mean value of the mesh area of the connected domain around the damaged position of the net includes:
[0079] The adjacent positions of the damaged net clothing position are selected according to the damaged net clothing position, and the average value of the mesh areas of the adjacent positions is used as the average value of the mesh areas of the connected domain surrounding the damaged net clothing position.
[0080] The embodiment of the present invention provides a method for identifying damage to deep-water nets, referring to Figure 2 , Figure 2 The figure is a flow chart of a first embodiment of a method for identifying damage to deep-water nets according to the present invention.
[0081] In this embodiment, the deep-water net damage identification method includes:
[0082] Step S1000: grayscale processing is performed on the collected original image of deep-water net to obtain a single-channel grayscale deep-water net image.
[0083] In this embodiment, the image acquisition device is bound to the robot, and moves with the robot underwater to collect the original image of deep-water nets at a specific position. Then, the collected original image of deep-water nets is gray-processed, and the original image of deep-water nets is transformed from RGB three channels into a single-channel gray-scale image according to a specific conversion rule, which reduces the calculation pressure for subsequent processing and improves the overall detection efficiency.
[0084] Step S2000: performing image hybrid enhancement processing on the single-channel grayscale deep-water net image to obtain a hybrid enhanced deep-water net image.
[0085] In this embodiment, the collected underwater net images often have more noise points due to the influence of various environmental factors, and the local contrast is too low, which is prone to false detection problems when performing damage detection on them. Therefore, it is necessary to enhance the underwater net images, that is, to perform image hybrid enhancement processing on the single-channel grayscale deep-water net image obtained after grayscale processing in step S10 to obtain a hybrid enhanced deep-water net image.
[0086] Step S3000: performing local adaptive threshold binarization processing on the hybrid enhanced deep-water net image to obtain a binarized deep-water net image.
[0087] In this embodiment, the hybrid enhanced deep-water net image obtained by the image hybrid enhancement processing in step S20 is binarized, the purpose of which is to separate the target from the background, in preparation for the subsequent damage detection of the net. The most commonly used method for grayscale image binarization is the threshold method, which uses the difference between the target and the background in the image to set the image to two different levels, select a suitable threshold to determine whether a pixel is the target or the background, and thus obtain a binary image. Most of the solutions in the prior art use a single threshold algorithm, that is, only one threshold is selected for the binarization process of an image, and "dark spots" will appear in areas with relatively low brightness and relatively high thresholds. In this embodiment, local adaptive threshold binarization is adopted, that is, the binarization threshold at the pixel position is determined according to the pixel value distribution of the pixel's neighborhood block. The advantage of this is that the binarization threshold at each pixel position is not fixed, but is determined by the distribution of the surrounding neighborhood pixels. The binarization threshold of the image area with higher brightness is usually higher, while the binarization threshold of the image area with lower brightness will be smaller accordingly. Local image areas with different brightness, contrast, and texture will have corresponding local binarization thresholds.
[0088] Step S4000: performing connected domain segmentation processing on the binary deep-water net image to obtain a mesh label sequence and a mesh area sequence.
[0089] The mesh sizes in the net under normal conditions are basically the same. If the net is damaged, the mesh at the damaged position will have a larger area abnormal value. Therefore, in this embodiment, it is necessary to first calculate the area of the mesh and attach a label value to each mesh. The local adaptive threshold binarization processing is performed in step S30, and the binary deep-water net image obtained is identified as a connected domain using the connected domain segmentation processing technology. The label value is marked from 1 until the last mesh connected domain n is marked. At the same time, small area noise points and incomplete edge meshes are ignored to obtain a mesh label sequence and a mesh area sequence. A label value is marked for each mesh, so that the damaged position of the net can be directly determined based on the mesh label.
[0090] Step S5000: performing unsupervised anomaly detection processing on the mesh area sequence, extracting abnormal values, and determining mesh damage recognition results according to mesh labels corresponding to the abnormal values.
[0091] In this embodiment, when performing outlier detection on the mesh area sequence, the commonly used means in the prior art is to first find the mesh area ascending sequence, and then find the area gradient sequence. Then, the abnormality of the mesh area characteristic gradient curve is judged by setting a threshold, and finally, the damage position index is returned to the original image, and the damage of the net is selected. There are two main problems with the outlier screening algorithm: one is that the calculation process is relatively complicated; the other is that it is difficult to obtain the conditional threshold, which needs to be obtained based on the characteristics of the net and breeding experience. HBOS (Histogram-based Outlier Score, unsupervised anomaly detection algorithm based on histogram) is an unsupervised anomaly detection algorithm with simple principle, fast calculation speed and generally very effective. Its basic assumption is that each dimension of the data set is independent of each other, and then each dimension is divided into intervals. The higher the density of the interval, the lower the anomaly score. The mesh size under normal conditions is basically the same, and the damaged mesh area is large, so outliers can be quickly extracted in the histogram of the mesh area sequence. HBOS is used to screen outliers in the mesh area sequence, and the damaged position and degree of the net are determined according to the corresponding mesh labels.
[0092] In this embodiment, in order to overcome the shortcomings of low local contrast of underwater net images and difficulty in extracting outliers during net damage detection, a net damage identification method based on underwater image enhancement and unsupervised anomaly detection is adopted to improve the accuracy and applicability of deep-water cage net damage detection. The underwater image enhancement technology is a "sandwich" hybrid enhancement technology for underwater net images, that is, guided filtering, contrast-limited adaptive histogram equalization, and guided filtering are performed on the image in the form of window sliding, and the local adaptive threshold binarization algorithm is used to perform binarization operations on the image, which effectively overcomes the problem of low local contrast of underwater images and reduces the probability of false detection of damaged meshes; at the same time, an unsupervised anomaly detection algorithm is introduced to simplify the net damage detection process and improve applicability.
[0093] Optionally, the step of performing image mixing enhancement processing on the single-channel grayscale deep-water net image comprises:
[0094] Step S2001: performing a guided filtering operation on the single-channel grayscale deep-water net image using a window image of a preset size;
[0095] Step S2002: using contrast-limited adaptive histogram equalization to perform a local contrast enhancement processing operation on the window image after the guided filtering operation;
[0096] Step S2003: performing filtering operation again on the window image after the local contrast enhancement operation with the window image of a preset size.
[0097] In this embodiment, during the image preprocessing process, due to the uneven underwater illumination, the image is prone to local contrast reduction, resulting in the loss of underwater net image details in subsequent processing. And because a single traditional image enhancement algorithm cannot achieve good results, a "sandwich" underwater net image hybrid enhancement technology is proposed, that is, guided filtering, contrast-limited adaptive histogram equalization, and guided filtering are performed on the image in the form of window sliding to perform image hybrid enhancement processing, including the following steps:
[0098] 1. Guided filtering: Select the guided filtering algorithm to filter the window image, denoising while retaining image details. Compared with traditional Gaussian filtering, guided filtering assumes a window w centered on a pixel k. k There is a local linear relationship in
[0099]
[0100] , where q is the output image, I is the guide image, i is the pixel index, and a k and b k is the invariant coefficient of the linear function when the window center is located at k, and the window size is selected as the preferred empirical value of 8×8. Based on the local linear relationship, the guided filter retains the edge details of the image to a greater extent while removing noise. Since the algorithm complexity of the guided filter is independent of the size of the window, it has more advantages in the efficiency of large-scale image processing than the bilateral filter. Since the object to be studied is the underwater net image, the details of the net wire need to be retained as much as possible, so the guided filter algorithm is selected to filter the image.
[0101] like Figure 3 Shown, a) original image; b) Gaussian filter effect; c) guided filter effect. The original image of the net clothing contains a lot of noise. Although the noise is significantly reduced after smoothing with the Gaussian filter, the whole image is blurred and the details of the network cable are not fully preserved. Compared with the Gaussian filter, the guided filter can better retain the details of the network cable while removing noise, and the filtered image is more transparent overall.
[0102] 2. Contrast Limited Adaptive Histogram Equalization: Use CLAHE (Contrast Limited Adaptive Histogram Equalization) to process the window image and enhance the local contrast. Due to the influence of factors such as shooting angle, lighting angle, and deformation of the net, the light intensity in the image is often unevenly distributed. The histogram components of the dark part of the image are concentrated at the lower grayscale end, while the histogram components of the bright part tend to the higher grayscale end. The grayscale distribution of the image is uneven, and its grayscale distribution is concentrated in a narrower range. The details of the image are not clear enough and the local contrast is low. If the grayscale image is directly binarized, the details of the dark part of the mesh will not be retained, which will cause false detection of subsequent damage detection, but the details of the bright part of the mesh will be well retained and will not be affected. Therefore, it is necessary to enhance the underwater net image to expand the grayscale range of the image or make the grayscale evenly distributed, thereby increasing the contrast and making the image details clear. CLAHE technology is used in this embodiment. Unlike traditional histogram equalization, boundary conditions are added to CLAHE to limit excessive amplification of the contrast.
[0103]
[0104] ,in, H k (bin) are the grayscale histograms of the window after CLAHE and traditional histogram equalization; L and T are the rising height and clipping threshold of the histogram, respectively. k The histogram of the contrast enhancement is obtained by redistributing the brightness. k (bin), and use adaptive threshold T to limit the histogram contrast to overcome the shortcomings of image discontinuity and over-enhancement. Therefore, CLAHE is used to process the grayscale image of underwater nets to overcome the influence of uneven illumination in the image, improve the local contrast of dark parts in the grayscale image, enhance the details of the net wires concerned in the image, and provide guarantee for the accuracy of subsequent damage detection.
[0105] like Figure 4 As shown in the figure, in the grayscale histogram of the original image, the grayscale values of the image are mainly concentrated in the interval (50, 150), and there is even no distribution in the low grayscale value and high grayscale value areas, and the pixels with two grayscale values that are too close are difficult to distinguish. After CLAHE processing, the grayscale histogram of the image is more evenly distributed compared to the original image, and the overall slope of the cumulative histogram is reduced. It can be understood that the grayscale histogram is more evenly distributed compared to the original image, the grayscale value gap between the two pixels is widened, and the two points are easier to distinguish, which means that the local contrast of the image is improved and the dark details can be displayed. This is more obvious in the binarization of the image.
[0106] 3. Guide filtering again: Repeat the filtering operation in step 1 to remove the local noise generated in step 2. Since CLAHE limits the histogram contrast, it is inevitable that image distortion and noise will occur in the actual processing process. Therefore, the image filtering and noise reduction process in step 1 is performed again. Note that only the image data in the window is processed at this time, and the subsequent sliding window will traverse the original image.
[0107] Optionally, the step of obtaining the hybrid enhanced deep-water net image comprises:
[0108] Step S2004: Slide the window image, and after each slide, perform the step of performing image hybrid enhancement processing on the single-channel grayscale deep-water net image until the single-channel grayscale deep-water net image is traversed to obtain the hybrid enhancement processed deep-water net image.
[0109] In this embodiment, starting from the upper left pixel point, after the "sandwich" mesh image hybrid enhancement processing, all subsequent pixels are processed in sequence from left to right and from top to bottom, and the "sandwich" mesh image hybrid enhancement processing is performed at each pixel point until the single-channel grayscale deep-water mesh image is traversed to obtain the hybrid enhancement processed deep-water mesh image. It should be noted that in this embodiment, only one traversal of the original image is required to obtain the final damage degree detection result.
[0110] Optionally, the step of performing local adaptive threshold binarization processing on the hybrid enhanced deep-water net image comprises:
[0111] The hybrid enhanced deep-water net image is binarized and inverted using a local adaptive threshold binarization algorithm, and the hybrid enhanced deep-water net image is simplified into a binary image, in which the background is the net wire and the target is the mesh.
[0112] In this embodiment, the net grayscale image (i.e., the hybrid enhanced deep-water net image) is subjected to a local adaptive threshold binarization operation and inverted, and the enhanced grayscale image is simplified into a binary image, with the background being the net wire and the target being the mesh. In the image preprocessing process, the local adaptive threshold binarization algorithm is used to perform a binarization operation on the image, improve the local contrast and enhance the details of the net wire in the image, and suppress the amplification of the image noise, which can effectively reduce the probability of false detection of damaged meshes.
[0113] from Figure 5 It can be clearly observed that the image detail retention is improved after image enhancement, especially the improvement of the image edge mesh details. However, after the unprocessed original image is binarized, the edge continuous mesh lines are discontinuous, which is easily misdetected as a damaged point in the subsequent damaged mesh detection, affecting the detection accuracy.
[0114] Optionally, the step of performing connected domain segmentation processing on the binary deep-water net image comprises:
[0115] Using an eight-neighborhood indexing method, each mesh is identified as a connected domain, and each connected domain is marked;
[0116] The number of pixels in each mesh is counted as the area of the corresponding mesh, and the meshes at the edge of the binary deep-water net image and the meshes whose area is smaller than the preset background area are set as the background.
[0117] In this embodiment, the connected domain segmentation process is performed on the binary deep-water net image using the 8-neighborhood indexing method. Each mesh is identified as a connected domain, and is marked starting from the region name (i.e., label value) of 1 until the last mesh connected domain n is marked. At the same time, the number of pixels in each mesh is counted as the area of each mesh, and the meshes with an area value less than 100 (i.e., the preset background area, which is a technical experience value) and the meshes at the edge of the net are set as the background. Finally, the mesh area of each connected domain is marked at the center of the connected domain, as shown in FIG. Figure 6 As shown in the figure, the mesh label sequence and mesh area sequence are obtained. Figure 6 Taking 506-159 in as an example, 506 is the mesh connected domain area, 159 is the mesh connected domain label, and the label is marked according to the area size, that is, 506 is the mesh connected domain area of the 159th size.
[0118] Optionally, the step of determining the net damage identification result according to the mesh label corresponding to the abnormal value includes:
[0119] Step S5001: determining the damaged position of the net according to the mesh label corresponding to the abnormal value, and calculating the average value of the mesh area of the connected domain around the damaged position of the net;
[0120] Step S5002: taking the ratio of the mesh area at the damaged position of the net to the average mesh area as the net damage detection result.
[0121] In this embodiment, if Figure 7 As shown in the figure, each mesh is divided into multiple intervals according to the area. The interval with fewer samples has a higher probability of being an outlier. It can be seen that the interval where the outlier point is located is separated from the interval on the left by a large gap, and there is only one point in this interval (note that the vertical axis is a logarithmic coordinate). The HBOS algorithm is used to filter out outliers in the mesh area sequence, and the location and degree of damage of the net are determined according to the corresponding mesh label.
[0122] In the traditional process of abnormal detection of damaged nets, it is generally necessary to set a threshold to extract outliers, but the threshold acquisition needs to be based on the characteristics of the nets and the breeding experience, and there is no clear standard, which limits the applicability of the damage detection algorithm. In this embodiment, an unsupervised anomaly detection algorithm based on a histogram is introduced—HBOS (Histogram-based Outlier Score). The core idea is to divide the samples into multiple intervals according to the characteristics, and the intervals with a small number of samples have a high probability of being outliers. According to the HBOS algorithm, the outliers of the net feature data can be quickly determined, and then the abnormal areas of the net can be identified, simplifying the underwater net damage detection process.
[0123] Optionally, the step of calculating the mean value of the mesh area of the connected domain around the damaged position of the net includes:
[0124] Step S5003: selecting adjacent positions of the damaged net clothing position according to the damaged net clothing position, and taking the average value of the mesh areas of the adjacent positions as the average value of the mesh areas of the connected domain surrounding the damaged net clothing position.
[0125] In this embodiment, the average mesh area of the connected domain around the damaged position of the net is calculated. That is, the average area of the four mesh connected domains adjacent to the damaged position of the net clothing is selected, and the degree of mesh damage R is the mesh area A at the damaged position i and The ratio of
[0126]
[0127] The final net damage detection results are as follows: Figure 8 As shown, the final damage degree in this embodiment is R=80.65.
[0128] In summary, 1. The embodiment of the present invention proposes a "sandwich" type underwater net image hybrid enhancement technology during image preprocessing, that is, guided filtering, contrast-limited adaptive histogram equalization, and guided filtering are performed on the image in the form of window sliding; at the same time, a local adaptive threshold binarization algorithm is used to perform a binarization operation on the image, which effectively overcomes the problem of low local contrast of underwater images and reduces the probability of false detection of damaged meshes;
[0129] 2. An unsupervised anomaly detection algorithm is introduced in the process of detecting the abnormality of net damage. The abnormal points of the net are directly judged through the mesh area histogram, without introducing the abnormal feature threshold of the net. This improves the applicability of the detection algorithm and simplifies the detection process.
[0130] In addition, an embodiment of the present invention further provides a deep-water net damage identification device, the deep-water net damage identification device comprising:
[0131] A grayscale processing module is used to perform grayscale processing on the collected original image of deep-water net to obtain a single-channel grayscale deep-water net image;
[0132] An image hybrid enhancement processing module is used to perform image hybrid enhancement processing on the single-channel grayscale deep-water net image to obtain a hybrid enhanced deep-water net image;
[0133] A local adaptive threshold binarization processing module is used to perform local adaptive threshold binarization processing on the hybrid enhanced deep-water net image to obtain a binarized deep-water net image;
[0134] A connected domain segmentation processing module is used to perform connected domain segmentation processing on the binary deep-water net image to obtain a mesh label sequence and a mesh area sequence;
[0135] The unsupervised anomaly detection processing module is used to perform unsupervised anomaly detection processing on the mesh area sequence, extract abnormal values, and determine the mesh damage recognition result according to the mesh labels corresponding to the abnormal values.
[0136] In addition, an embodiment of the present invention also provides a deep-water net damage identification device, which includes: a memory, a processor, and a deep-water net damage identification program stored in the memory and executable on the processor, and when the deep-water net damage identification program is executed by the processor, the steps of the deep-water net damage identification method as described above are implemented.
[0137] In addition, an embodiment of the present invention further provides a computer-readable storage medium, on which a deep-water net damage identification program is stored. When the deep-water net damage identification program is executed by a processor, the steps of the deep-water net damage identification method as described above are implemented.
[0138] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system 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 system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.
[0139] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.
[0140] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.
[0141] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A method for identifying damaged deep-water nets, characterized in that: The deep-water net damage identification method comprises the following steps: The collected original image of deep-water net is gray-processed to obtain a single-channel gray-scale deep-water net image; Performing image hybrid enhancement processing on the single-channel grayscale deep-water net image to obtain a hybrid enhanced deep-water net image; Performing local adaptive threshold binarization processing on the hybrid enhanced deep-water net image to obtain a binarized deep-water net image; Performing connected domain segmentation processing on the binary deep-water net image to obtain a mesh label sequence and a mesh area sequence; Performing unsupervised anomaly detection processing on the mesh area sequence, extracting outliers, and determining mesh damage recognition results according to mesh labels corresponding to the outliers; The step of performing image mixing enhancement processing on the single-channel grayscale deep-water net image comprises: Performing a guided filtering operation on the single-channel grayscale deep-water net image using a window image of a preset size; Using limited contrast adaptive histogram equalization to perform a local contrast enhancement processing operation on the window image after the guided filtering operation; Performing filtering operation again on the window image after the local contrast enhancement operation with the window image of a preset size; The step of performing local adaptive threshold binarization processing on the hybrid enhanced deep-water net image comprises: The hybrid enhanced deep-water net image is binarized and inverted using a local adaptive threshold binarization algorithm, and the hybrid enhanced deep-water net image is simplified into a binary image, in which the background is the net wire and the target is the mesh.
2. The deep-water net damage identification method according to claim 1, characterized in that: The step of obtaining the hybrid enhanced deep-water net image comprises: Slide the window image, and after each slide, perform the step of performing image mixed enhancement processing on the single-channel grayscale deep-water net image, until the single-channel grayscale deep-water net image is traversed to obtain the mixed enhancement processed deep-water net image.
3. The method for identifying damage to deep-water nets according to claim 1, characterized in that: The step of performing connected domain segmentation processing on the binary deep-water net image comprises: Using an eight-neighborhood indexing method, each mesh is identified as a connected domain, and each connected domain is marked; The number of pixels in each mesh is counted as the area of the corresponding mesh, and the meshes at the edge of the binary deep-water net image and the meshes whose area is smaller than the preset background area are set as the background.
4. The method for identifying damage to deep-water nets according to claim 1, characterized in that: The step of determining the net damage identification result according to the mesh label corresponding to the abnormal value comprises: Determine the damaged position of the net according to the mesh label corresponding to the abnormal value, and calculate the average value of the mesh area of the connected domain around the damaged position of the net; The ratio of the mesh area at the damaged position of the net to the average mesh area is taken as the net damage detection result.
5. The method for identifying damage to deep-water nets according to claim 4, characterized in that: The step of calculating the mean value of the mesh area of the connected domain around the damaged position of the net includes: The adjacent positions of the damaged net clothing position are selected according to the damaged net clothing position, and the average value of the mesh areas of the adjacent positions is used as the average value of the mesh areas of the connected domain surrounding the damaged net clothing position.
6. A deep-water net damage identification device, characterized in that: The deep-water net damage identification device comprises: A grayscale processing module is used to perform grayscale processing on the collected original image of deep-water net to obtain a single-channel grayscale deep-water net image; An image hybrid enhancement processing module is used to perform image hybrid enhancement processing on the single-channel grayscale deep-water net image to obtain a hybrid enhanced deep-water net image; A local adaptive threshold binarization processing module is used to perform local adaptive threshold binarization processing on the hybrid enhanced deep-water net image to obtain a binarized deep-water net image; A connected domain segmentation processing module is used to perform connected domain segmentation processing on the binary deep-water net image to obtain a mesh label sequence and a mesh area sequence; An unsupervised anomaly detection processing module is used to perform unsupervised anomaly detection processing on the mesh area sequence, extract outliers, and determine a mesh damage recognition result according to mesh labels corresponding to the outliers; The image mixing enhancement processing module is also used for: Performing a guided filtering operation on the single-channel grayscale deep-water net image using a window image of a preset size; Using limited contrast adaptive histogram equalization to perform a local contrast enhancement processing operation on the window image after the guided filtering operation; Performing filtering operation again on the window image after the local contrast enhancement operation with the window image of a preset size; The local adaptive threshold binarization processing module is also used for: The hybrid enhanced deep-water net image is binarized and inverted using a local adaptive threshold binarization algorithm, and the hybrid enhanced deep-water net image is simplified into a binary image, in which the background is the net wire and the target is the mesh.
7. A deep-water net damage identification device, characterized in that: The deep-water net damage identification device comprises: a memory, a processor and a deep-water net damage identification program stored in the memory and executable on the processor, wherein the deep-water net damage identification program is configured to implement the steps of the deep-water net damage identification method as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a deep-water net damage identification program, which, when executed by a processor, implements the steps of the deep-water net damage identification method according to any one of claims 1 to 5.
