Net cage culture underwater netting damage identification method

Abstract

The invention discloses a cage culture underwater netting damage identification method. According to the method, through a netting damage detection module carried by an underwater robot, a semantic segmentation model is utilized to segment a netting image, and a segmentation result is optimized in combination with image post-processing, so that a netting damage area is accurately identified; when it is detected that the netting is damaged, the underwater robot slowly gets close to a normal netting near a netting damage area, and a marker is fixed to the netting through an elastic clamp of the netting damage marking device; a stroboscopic signal lamp is arranged on the marker, so that the diver can be helped to quickly find out the damaged netting. The device has the advantages that the netting damage detection accuracy is high, the marking device is simple in structure and can be recycled for multiple times, the problem that the netting damage is difficult to find quickly when a diver maintains the netting is effectively solved, and the intelligent level of cage culture is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of cage aquaculture engineering and equipment, and particularly relates to a method for identifying underwater net clothing breakage in cage aquaculture. Background Art

[0002] With the progress of technology and the development of fishery, cage aquaculture has become a key measure to relieve the pressure of fish farming and break through the problems of resource limitation and space constraint. As a key component of the cage, the net clothing may be damaged to varying degrees during the aquaculture process due to factors such as ocean current impact, fish hitting or biting the net, material aging, biological attachment, and net clothing cleaning. In severe cases, it may lead to the escape of cultured fish, thus affecting the aquaculture benefit and harming the marine ecological environment. Therefore, it is particularly important to regularly inspect the underwater net clothing, accurately detect and locate the breakage of the net clothing, so as to carry out maintenance in a timely manner.

[0003] At present, the inspection of underwater net clothing in cage aquaculture mainly relies on manual visual inspection, including direct inspection by divers and remote control inspection by underwater robots. With the rapid development of technologies such as sensors, artificial intelligence, and robots, and the increasing demand for intelligent cage aquaculture, the development of intelligent algorithms and devices for net clothing inspection that can be used by underwater robots has become a trend. In the actual cage aquaculture environment, due to the complex underwater environment and limited operating conditions, the existing methods for detecting net clothing breakage still have many deficiencies and are difficult to meet the application requirements. In addition, due to the dynamic and biological characteristics of the cage aquaculture environment, the existing positioning methods have errors when determining the position of the underwater robot in the cage. This results in the difficulty for the underwater robot to determine the position of the net clothing breakage on the cage based on its own position information when detecting the breakage of the net clothing, thus increasing the difficulty of net clothing maintenance. At present stage, net clothing maintenance mainly relies on divers to carry out underwater operations. However, when divers are working underwater, they are easily affected by factors such as low visibility, ocean current, cultured fish, and biological attachment, resulting in the difficulty to quickly find the net clothing breakage even when the approximate position of the net clothing breakage is known, greatly increasing the safety risk of diving operations. Therefore, it is urgent to develop a method for identifying underwater net clothing breakage in cage aquaculture to provide effective means for detecting and marking net clothing breakage for underwater robots during cage inspection. Summary of the Invention

[0004] The main purpose of the present invention is to provide a method for identifying underwater net clothing breakage in cage aquaculture, aiming to provide effective means for detecting and marking net clothing breakage for underwater robots during cage inspection.

[0005] The technical solution adopted by the present invention is as follows:

[0006] I. A method for identifying underwater net clothing breakage in cage aquaculture

[0007] The described method for identifying damage to the underwater net of cage aquaculture includes the following steps:

[0008] S100) Use the net damage detection module carried on the underwater robot to perform real-time detection on the net images collected by the underwater robot, and obtain the detection result.

[0009] Specifically, in step S100, the net damage detection module includes a semantic segmentation unit and an image post-processing unit; the semantic segmentation unit is used to segment the net images collected by the underwater robot using a semantic segmentation model to obtain a corresponding segmentation mask, and several areas of suspected net damage are marked in the segmentation mask; the image post-processing unit is used to process the segmentation mask, obtain the detection result and output the detection result to the underwater robot; the detection result includes no net damage and net damage. If the detection result is net damage, the detection result also includes the position of the net damage area.

[0010] Preferably, in the net damage detection module, the semantic segmentation model uses the SegFormer model.

[0011] Optionally, in the net damage detection module, the construction process of the semantic segmentation model for net image segmentation is specifically as follows:

[0012] S1) Pre-acquire several net images, annotate each net image to obtain the corresponding segmentation mask for each net image. Each net image and its corresponding segmentation mask form an image-mask pair. All image-mask pairs constitute the net image segmentation data set, and the net image segmentation data set is divided into a training set, a validation set, and a test set;

[0013] S2) Use the training set and validation set in the net image segmentation data set to train at least one semantic segmentation model to obtain the optimal model weights corresponding to each semantic segmentation model. Use the test set in the net image segmentation data set to test various semantic segmentation models respectively, and select the optimal model from all semantic segmentation models as the semantic segmentation model for net image segmentation according to the test results.

[0014] The process of the image post - processing unit processing the segmentation mask to obtain the detection result includes the following steps: First, remove the to - be - confirmed net - clothing damage areas in the segmentation mask where the pixel area is lower than the preset threshold, and fill in each to - be - confirmed net - clothing damage area that does not fill the corresponding mesh; Subsequently, detect whether there is net - clothing damage in each to - be - confirmed net - clothing damage area in the segmentation mask: If there is net - clothing damage in the to - be - confirmed net - clothing damage area, mark the to - be - confirmed net - clothing damage area as a net - clothing damage area; If there is no net - clothing damage in the to - be - confirmed net - clothing damage area, remove the to - be - confirmed net - clothing damage area; Finally, if the segmentation mask does not contain a net - clothing damage area, output the detection result of no net - clothing damage; If the segmentation mask contains a net - clothing damage area, output the detection result of net - clothing damage and the positions of each net - clothing damage area.

[0015] In a specific implementation, the removal of the to - be - confirmed net - clothing damage area is specifically: changing the pixel values of the to - be - confirmed net - clothing damage area to the pixel values of the background area.

[0016] Specifically, the process of detecting whether there is net - clothing damage in each to - be - confirmed net - clothing damage area in the segmentation mask includes the following steps: Obtain the Euclidean distance between the center - point coordinates of each mesh in the segmentation mask and the center - point coordinates of the to - be - confirmed net - clothing damage area, and screen out the meshes within the preset threshold distance as the adjacent meshes of the to - be - confirmed net - clothing damage area; Obtain the pixel area of each adjacent mesh, screen out the adjacent meshes with too large or too small pixel areas according to the preset pixel - area threshold, and obtain the average value of the pixel areas of the remaining adjacent meshes; Obtain the ratio between the pixel area of the to - be - confirmed net - clothing damage area and the average value of the pixel areas of the remaining adjacent meshes. If the ratio is greater than or equal to 2, there is net - clothing damage in the to - be - confirmed net - clothing damage area. If the ratio is less than 2, there is no net - clothing damage in the to - be - confirmed net - clothing damage area.

[0017] S200) When the underwater robot does not detect net - clothing damage, go to step S500; When the underwater robot detects net - clothing damage, the underwater robot adjusts its position relative to the net - clothing damage area and approaches the normal net - clothing near the net - clothing damage area until the jaws of the elastic clamp of the marker of the net - clothing damage marking device pass through the normal net - clothing near the net - clothing damage area.

[0018] S300) After the jaws of the elastic clamp of the marker of the net - clothing damage marking device pass through the normal net - clothing near the net - clothing damage area, the underwater robot controls the controller of the net - clothing damage marking device to mark the net - clothing damage and fix the marker on the normal net - clothing near the net - clothing damage area.

[0019] The specific steps of step S300 are as follows: When the jaws of the elastic clamp of the marker of the netting damage marking device pass through the normal netting near the netting damage area, the underwater robot controls the waterproof servo of the controller of the netting damage marking device to rotate the front end of the long buckle back to the horizontal position; at this time, the built-in spring of the elastic clamp takes effect and pushes the marker forward to eject; at the same time, the jaws of the elastic clamp quickly clamp under the action of elastic force to fix the marker on the normal netting near the netting damage area, and the strobe light on the marker strobes at a preset certain frequency.

[0020] S400) The underwater robot moves away from the netting to complete the marking of the netting damage;

[0021] S500) The underwater robot stops the monitoring operation, or after moving to the next monitoring point, performs the operation according to the same process as steps S100 to S400.

[0022] II. A netting damage marking device applied to the above-mentioned underwater netting damage marking method for cage aquaculture

[0023] The netting damage marking device is installed at the bottom of the underwater robot and includes a marker, a controller, and a mounting chassis; the mounting chassis is fixedly installed on the underwater robot, the front end of the mounting chassis is fixedly connected to the rear end of the controller, the front end of the controller is detachably connected to the marker, and the controller is electrically connected to the control module of the underwater robot;

[0024] The marker includes a buckle base, an upper shell, an elastic clamp, a lower shell, and a strobe light; the buckle base is fixedly installed on the top surface of the upper shell, the upper shell and the lower shell are arranged up and down to form a cavity structure, the elastic clamp is arranged in the cavity structure, the jaws of the elastic clamp face forward and the front end extends out of the cavity structure; the strobe light is installed on the lower shell;

[0025] The controller includes a main body bracket, a waterproof servo, and a long buckle; the waterproof servo is arranged inside the main body bracket, the steering wheel of the waterproof servo is fixedly connected to the rear end of the long buckle, the front end of the long buckle is detachably buckled to the buckle base on the marker; a limiting rod is installed on the front end face of the main body bracket, and the limiting rod extends into the marker and is arranged in cooperation with the elastic clamp.

[0026] The detachable snap - fit connection between the front end of the long snap and the snap - female base on the marker specifically means that: a convex portion is provided at the front end of the long snap, so that the front end of the long snap forms an "L - shaped" structure. When the front end of the long snap is in the horizontal position, the front end of the long snap can slide back and forth in the through - slot of the snap - female base. At this time, insert the front end of the long snap into the through - slot of the snap - female base, and then adjust the convex portion to the vertical position. The convex portion abuts against the front end of the snap - female base, so that the long snap and the snap - female base form an interlocking structure. When the convex portion is adjusted to the horizontal position again, the long snap disengages from the snap - female base in the front - to - back direction, thereby releasing the interlocking structure.

[0027] The elastic clamp includes a pliers - shaped link mechanism, two elastic support components symmetrically arranged on the upper and lower sides of the pliers - shaped link mechanism, a slider for realizing the transmission connection between the pliers - shaped link mechanism and the elastic support components, and a number of short cylinders arranged between the two elastic support components;

[0028] The pliers - shaped link mechanism adopts a rhombic four - link mechanism. The ends of the rhombic four - link mechanism are all hinged to the slider, and barbs are provided at the front ends.

[0029] Each elastic support component mainly consists of a metal skeleton, a skeleton, a spring block, and a spring; the skeleton is stacked on the inner side (i.e., the side away from the pliers - shaped link mechanism) of the metal skeleton. A strip - shaped slot arranged in the front - to - back direction is opened in the center of the skeleton, and the slider is slidably arranged in the strip - shaped slot; the spring block is installed at the front end of the skeleton, and the spring is arranged between the spring block and the slider. The two ends of the spring respectively abut against the spring block and the slider; the metal skeletons in the two elastic support components are connected by a number of short cylinders;

[0030] When the front end of the limit rod extends into the elastic clamp, the front end face of the limit rod contacts the rear end face of the slider. The limit rod can drive the slider to move forward and approach the spring block, opening the jaws and compressing the spring;

[0031] When the front end of the limit rod disengages from the elastic clamp, the restoring force of the spring drives the slider to move backward and away from the spring block, closing the jaws.

[0032] Specifically, the rhombic four - link mechanism mainly consists of two support rods and two clamping linkages. The two clamping linkages are cross - arranged and are hinged by a connecting rotating shaft. The connecting rotating shaft is located at the intersection of the two clamping linkages, so that a jaw is formed between the front ends of the two clamping linkages. The ends of the two clamping linkages are respectively hinged to the front ends of the two support rods, and the ends of the two support rods are both hinged to the slider. Barbs are provided on the inner sides of the front ends of the two clamping linkages.

[0033] Furthermore, each elastic support component further includes a gasket and a bearing. The bearing is arranged at the front end of the bone frame, and the bearing is located on the front side of the strip-shaped groove. The end of the connecting rotating shaft of the rhombic four-link mechanism extends out of the clamping connecting rod and then extends into the bearing, and a gasket is arranged between the bearing and the clamping connecting rod.

[0034] III. An underwater robot

[0035] The underwater robot integrates the net clothing damage detection module and the net clothing damage marking device as described above.

[0036] Based on the detection of net clothing damage, the present invention fixes the marker on the normal net clothing near the net clothing damage area through an elastic clamp, and uses a stroboscopic signal lamp to help the diver quickly find the net clothing damage, effectively meeting the requirements of net clothing damage detection, positioning and maintenance in net cage aquaculture.

[0037] The beneficial effects of the present invention are as follows:

[0038] 1) The present invention uses a semantic segmentation model to segment the net clothing image and combines a post-processing algorithm to optimize the semantic segmentation result, so as to more accurately identify the net clothing damage;

[0039] 2) The present invention utilizes the structural characteristics of the net clothing and fixes the marker on the net clothing through an elastic clamp;

[0040] 3) The stroboscopic signal lamp on the marker in the present invention can help the diver quickly find the net clothing damage;

[0041] 4) The net clothing damage marking device in the present invention has a simple structure, does not require an additional control circuit, and can be recycled and used multiple times. Description of the drawings

[0042] Figure 1 is a schematic flow chart of the net clothing damage marking of the underwater robot of the present invention;

[0043] Figure 2 is a schematic flow chart of the net clothing damage detection of the present invention;

[0044] Figure 3 is the original image and the annotated image of a partial net clothing image of the net clothing damage detection method of the present invention;

[0045] Figure 4 is an example diagram of the net clothing damage detection method of the present invention;

[0046] Figure 5 is a schematic flow chart of the image post-processing of the net clothing damage detection method of the present invention;

[0047] Figure 6 is a schematic structural diagram of the net clothing damage marking device of the underwater robot of the present invention;

[0048] Figure 7 is a schematic structural diagram of the netting damage marking device of the present invention;

[0049] Figure 8 is a schematic structural diagram of the marker of the present invention;

[0050] Figure 9 is a schematic structural diagram of the controller of the present invention;

[0051] Figure 10 is a schematic structural diagram of the elastic clamp of the present invention;

[0052] Figure 11 is a schematic principle diagram of the netting damage marking device of the present invention;

[0053] Figure 12 is a schematic diagram of the netting damage marking of the underwater robot of the present invention.

[0054] In the figure: 1. Underwater robot, 2. Netting damage marking device, 21. Marker, 22. Controller, 23. Installation chassis, 211. Snap female seat, 212. Upper shell, 213. Elastic clamp, 214. Lower shell, 215. Strobe signal lamp, 221. Main body bracket, 222. Waterproof servo, 223. Long snap, 224. Protective shell, 2131. Pincer link mechanism, 2132. Slide block, 2133. Metal skeleton, 2134. Skeleton, 2135. Spacer, 2136. Bearing, 2137. Spring block, 2138. Spring, 2139. Short cylinder. Specific embodiments

[0055] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0056] As Figure 1 shown, the method for identifying the damage of the underwater netting in the cage culture of the present invention includes the following steps:

[0057] S100) Use the camera device mounted on the underwater robot to collect the netting image at the current monitoring point. After the underwater robot obtains the netting image, use the netting damage detection module mounted on the underwater robot to perform real-time detection on the netting image collected by the underwater robot to obtain the detection result.

[0058] In step S100, the netting damage detection module includes a semantic segmentation unit and an image post-processing unit. The semantic segmentation unit uses a semantic segmentation model to segment the netting image collected by the underwater robot, and then obtains a segmentation mask, and several regions to be confirmed for netting damage are marked in the segmentation mask.

[0059] As Figure 2As shown, the construction process of the semantic segmentation model for netting image segmentation is specifically as follows:

[0060] S1) Obtain a number of netting images in advance, annotate each netting image to obtain the corresponding segmentation mask for each netting image. Each netting image and the corresponding segmentation mask form an image-mask pair. All image-mask pairs form the netting image segmentation dataset, which is divided into a training set, a validation set, and a test set.

[0061] The annotation is specifically as follows: The netting image is divided into a foreground region, a background region, and a netting damage region. Among them, the foreground region includes objects such as netting, biological attachment, cultured fish, and underwater robot cables, and the background region mainly includes seawater.

[0062] In the embodiment of the present invention, we annotated the netting images collected from the public dataset, the netting images obtained under laboratory conditions, and the netting images obtained in the actual cage culture environment. The pixels of each netting image were divided into a foreground region including netting, biological attachment, cultured fish, and underwater robot cables, a background region mainly composed of seawater, and a netting damage region.

[0063] In the embodiment of the present invention, the annotation method is specifically as follows: Change the pixel value of the foreground region to 0, change the pixel value of the background region to 1, and change the pixel value of the netting damage region to 2.

[0064] Figure 3 are the original image and the annotated image of part of the netting images. The original image and the annotation Figure 1 correspond one by one to form the netting image segmentation dataset, and are divided into a training set, a validation set, and a test set according to the ratio of 7:1.5:1.5.

[0065] S2) Use the training set and the validation set to train at least one semantic segmentation model to obtain the corresponding best model weights for each semantic segmentation model. Use the test set to test each semantic segmentation model, and select the optimal model from all semantic segmentation models as the semantic segmentation model for netting image segmentation according to the test results.

[0066] In the embodiment of the present invention, to improve the generalization ability of the semantic segmentation model, in the preprocessing of the semantic segmentation model, operations such as random blur, random brightness, random chromaticity, random contrast, random sharpness, and random RGB channel order change are added. As the preprocessing operation of the semantic segmentation model, new netting images and corresponding annotation images are generated according to the original images and annotation images of the netting images in the training set in each iteration of the model training.

[0067] On this basis, semantic segmentation models such as UNet, DeepLabV3+, PSPNet, KNet, SegFormer, Mask2Former, and FastSCNN were mainly trained for 20,000 iterations. Every 200 iterations, the mIoU, mAcc of the model parameters obtained from the training on the validation set were evaluated, as well as the IoU and Acc of each category. Finally, the best model weights of each semantic segmentation model were obtained. Each semantic segmentation model was tested using the test set. The test results showed that SegFormer performed best in terms of comprehensive performance. Therefore, SegFormer was selected as the semantic segmentation model for netting image segmentation.

[0068] S3) In step S100, the segmentation mask obtained by segmenting the netting image using the semantic segmentation model is used for image post-processing. The image post-processing unit is used to process the segmentation mask, obtain the detection result, and output the detection result to the control end of the underwater robot. The detection result includes that there is no netting breakage and that there is netting breakage. If the detection result is that there is netting breakage, the detection result also includes the position of the netting breakage area.

[0069] In the embodiment of the present invention, the trained semantic segmentation model is used to segment the netting image. As Figure 4 shown, since it is difficult for the segmentation accuracy of the semantic segmentation model to reach 100%, there may be a phenomenon of misidentification in the netting breakage area in the obtained segmentation mask. To avoid this situation, image post-processing is added on the basis of the prediction of the semantic segmentation model to improve the accuracy of netting breakage detection.

[0070] As Figure 2 and Figure 5 shown, the process of performing image post-processing on each segmentation mask specifically includes the following steps:

[0071] T1) Remove the to-be-confirmed netting breakage areas in the segmentation mask with a pixel area lower than the preset threshold, and fill in the to-be-confirmed netting breakage areas that do not fill the corresponding mesh.

[0072] In the embodiment of the present invention, by calculating the pixel areas of all to-be-confirmed netting breakage areas, the to-be-confirmed netting breakage areas with a pixel area less than 9 are removed. For each to-be-confirmed netting breakage area that does not fill the corresponding mesh, the pixel values of the background area in the mesh are changed to the pixel values of the netting breakage area.

[0073] T2) For the to-be-confirmed netting breakage areas, calculate the Euclidean distance between the center point coordinates of each mesh in the segmentation mask and the center point coordinates of the to-be-confirmed netting breakage areas, and screen out the meshes within the preset threshold as the adjacent meshes of the to-be-confirmed netting breakage areas.

[0074] In an embodiment of the present invention, for each area of the fishing net to be confirmed as damaged, the Euclidean distance between the center point coordinates of all mesh holes in the segmentation mask and the center point coordinates of the area of the fishing net to be confirmed as damaged is calculated, and they are sorted. The mesh holes with the top 20% of the distances are used as the adjacent mesh holes of the area of the fishing net to be confirmed as damaged.

[0075] T3) Calculate the pixel area of each adjacent mesh hole, filter out the adjacent mesh holes with too large or too small pixel areas, obtain the average value of the pixel areas of the remaining adjacent mesh holes, calculate the ratio between the pixel area of the area of the fishing net to be confirmed as damaged and this average value, and determine whether the area of the fishing net to be confirmed as damaged has fishing net damage through the ratio.

[0076] In an embodiment of the present invention, the adjacent mesh holes of the area of the fishing net to be confirmed as damaged are sorted according to the pixel area, the adjacent mesh holes with the top 40% and the bottom 80% of the pixel areas are filtered out, the average value of the pixel areas of the remaining adjacent mesh holes is taken, the ratio between the pixel area of the area of the fishing net to be confirmed as damaged and this average value is calculated. If the ratio is greater than or equal to 2, the area of the fishing net to be confirmed as damaged has fishing net damage. If the ratio is less than 2, the area of the fishing net to be confirmed as damaged has no fishing net damage, and there is a phenomenon of misidentifying the damaged area of the fishing net. The pixel value of the area of the fishing net to be confirmed as damaged is changed to the pixel value of the background area.

[0077] T4) Repeat the operations in steps T2) to T3) for other areas of the fishing net to be confirmed as damaged until all areas of the fishing net to be confirmed as damaged are traversed, and output the final detection result, as shown in the output image in Figure 4 as shown.

[0078] Through actual testing, the fishing net damage detection method based on SegFormer and image post - processing has an identification accuracy rate of more than 96.8% on the test set of the fishing net image segmentation dataset constructed for damage degrees greater than or equal to two normal mesh holes.

[0079] S200) When the underwater robot does not detect fishing net damage, go to step S500; when the underwater robot detects fishing net damage, the underwater robot adjusts its position relative to the damaged area of the fishing net and slowly approaches the normal fishing net near the damaged area of the fishing net until the jaws of the elastic clamp of the marker of the fishing net damage marking device pass through the normal fishing net near the damaged area of the fishing net.

[0080] Step S200 specifically includes the following steps: The control end of the underwater robot receives the detection result; when the detection result is that there is no netting breakage, operate according to step S500; when the detection result is that there is a netting breakage, the underwater robot adjusts its position relative to the netting breakage area according to the position of the netting breakage area, so that the jaws of the elastic clamp of the marker of the netting breakage marking device are aligned with the normal netting near the netting breakage area, and slowly approaches the normal netting near the netting breakage area until the jaws of the elastic clamp of the marker of the netting breakage marking device pass through the normal netting near the netting breakage area.

[0081] Specifically, the normal netting near the netting breakage area refers to: starting from the edge of the netting breakage area, extending outward by at least 1 mesh to a range of multiple meshes, the complete and unbroken netting. The extension distance range is at least to cover 1 complete mesh outside the breakage edge to ensure that the elastic clamp of the marker can firmly hold the normal netting and avoid marking failure caused by the loosening of the breakage edge. The specific distance can be dynamically adjusted according to the actual scenario (such as mesh size, netting material, breakage degree, etc.). For example, if the mesh diameter is large, the extension distance is correspondingly increased (such as extending 2 times the mesh diameter). If the breakage area is large or the netting tension is high, it needs to be extended to multiple meshes (such as 2 - 3 meshes) to provide sufficient clamping stability and operation space.

[0082] In some embodiments, during the process of the underwater robot adjusting its position relative to the netting breakage area according to the position of the netting breakage area, the underwater robot obtains the distance and angle between itself and the netting, makes the body attitude parallel to the normal vector of the netting surface, and at the same time makes the jaws of the elastic clamp of the marker of the netting breakage marking device align with the normal netting near the netting breakage area, and slowly approaches the normal netting near the netting breakage area until the jaws of the elastic clamp of the marker of the netting breakage marking device pass through the normal netting near the netting breakage area.

[0083] In some embodiments, the underwater robot can adjust its position relative to the netting breakage area according to the position of the netting breakage area through a vision system.

[0084] S300) After the jaws of the elastic clamp of the marker of the netting breakage marking device pass through the normal netting near the netting breakage area, the underwater robot controls the controller of the netting breakage marking device to mark the netting breakage, as Figure 11 shown.

[0085] Step S300 specifically includes the following steps: After the jaws of the elastic clamp of the net damage marking device pass through the normal net near the net damage area, the underwater robot controls the waterproof servo of the controller of the net damage marking device to rotate the front end of the long buckle back to the horizontal position. At this time, the built-in spring of the elastic clamp takes effect and pushes the marker forward to eject. At the same time, the jaws of the elastic clamp quickly clamp under the action of elastic force, fixing the marker on the normal net near the net damage area, and the strobe light on the marker flashes at a certain frequency.

[0086] S400) The underwater robot slowly moves away from the net to complete the marking of the net damage, as Figure 12 shown.

[0087] S500) The underwater robot stops the monitoring operation, or moves to the next monitoring point, and takes the next monitoring point as the current monitoring point, and operates according to the same process as steps S100 to S400.

[0088] The second aspect of the present invention provides a net damage marking device applied to the above-mentioned underwater net damage identification method for cage aquaculture.

[0089] As Figure 6 and Figure 7 shown, the net damage marking device 2 is installed at the bottom of the underwater robot 1, and includes a marker 21, a controller 22 and a mounting chassis 23. The mounting chassis 23 is fixedly installed on the underwater robot 1, the mounting chassis 23 is arranged forward, the front end of the mounting chassis is fixedly connected to the rear end of the controller 22, the front end of the controller 22 is detachably connected to the marker 21, the controller 22 is electrically connected to the control module of the underwater robot 1, and is controlled by the underwater robot 1.

[0090] As Figure 8 shown, the marker 21 includes a buckle socket 211, an upper housing 212, an elastic clamp 213, a lower housing 214 and a strobe light 215; a buckle socket 211 is fixedly installed on the top surface of the upper housing 212 near the controller 22, the upper housing 212 and the lower housing 214 are arranged up and down to form a cavity structure with openings at the front end and the rear end, an elastic clamp 213 is arranged in the cavity structure, and the jaws of the elastic clamp 213 extend out of the cavity structure and are arranged forward; a strobe light 215 is installed on the bottom surface of the lower housing 214.

[0091] As Figure 9As shown, the controller 22 includes a main body bracket 221, a waterproof servo 222, a long buckle 223, and a protective shell 224; the waterproof servo 222 is arranged inside the main body bracket 221, the steering wheel of the waterproof servo 222 is fixedly connected to the rear end of the long buckle 223, and the front end of the long buckle 223 is detachably snap-connected to the female buckle seat 211 on the marker 21. A limit rod is installed on the front end face of the main body bracket 221, and the limit rod extends into the marker 21 and is arranged in cooperation with the elastic clamp 213.

[0092] Furthermore, a protective shell 224 is arranged on each of the left and right sides of the main body bracket 221.

[0093] As Figure 10 shown, the elastic clamp 213 includes a pliers-link mechanism 2131, two elastic support components symmetrically arranged on the upper and lower sides of the pliers-link mechanism 2131, a slider 2132 for realizing the transmission connection between the pliers-link mechanism 2131 and the elastic support components, and a number of short cylinders 2139 arranged between the two elastic support components; the pliers-link mechanism 2131 adopts a rhombic four-link mechanism, the ends of the rhombic four-link mechanism are hinged to the slider 2132, and barbs are provided at the front ends, which can effectively prevent the marker 21 from falling off the fishing net.

[0094] Each elastic support component is mainly composed of a metal skeleton 2133, a skeleton 2134, a spring block 2137, and a spring 2138; the skeleton 2134 is stacked on the inner side (i.e., the side away from the pliers-link mechanism 2131) of the metal skeleton 2133, a strip-shaped groove is provided in the center of the skeleton 2134 and arranged in the front-rear direction, and the slider 2132 is slidably arranged in the strip-shaped groove; a spring block 2137 is installed at the front end of the skeleton 2134, a spring 2138 is arranged between the spring block 2137 and the slider 2132, and the two ends of the spring 2138 respectively abut against the spring block 2137 and the slider 2132.

[0095] On this basis, the process of the limit rod extending into the marker 21 and being arranged in cooperation with the elastic clamp 213 is specifically as follows: when the front end of the limit rod extends into the elastic clamp 213, the front end face of the limit rod contacts the rear end face of the slider 2132. By controlling the limit rod to continue to extend into the elastic clamp 213, the slider 2132 can be driven to move forward and approach the spring block 2137. When the marker 21 is connected to the controller 22, the slider 2132 is fixed at the front end of the strip-shaped groove through the limit rod, the jaws are opened and the spring 2138 is compressed, and at this time the spring 2138 is in a compressed state; when the marker 21 is separated from the controller 22, the front end of the limit rod disengages from the elastic clamp 213, and the restoring force of the spring 2138 drives the slider 2132 to move towards the rear end of the strip-shaped groove and away from the spring block 2137, so that the jaws clamp the fishing net.

[0096] Further, the rhombic four-bar linkage mechanism mainly consists of two support rods and two clamping linkages. The two clamping linkages are arranged crosswise and are hinged and connected through a connecting rotating shaft. The connecting rotating shaft is located at the intersection of the two clamping linkages, so that a jaw is formed between the front ends of the two clamping linkages. The ends of the two clamping linkages are respectively hinged and connected to the front ends of the two support rods, and the ends of the two support rods are both hinged and connected to the slider 2132. Barbs are provided on the inner sides of the front ends of the two clamping linkages. A through hole is provided at the front end of the bone frame 2134, a bearing 2136 is arranged in the through hole, the end of the connecting rotating shaft extends out of the clamping linkage and then extends into the bearing 2136, and a gasket 2135 is arranged between the bearing 2136 and the clamping linkage.

[0097] Further, the metal bone frames 2133 in the two elastic support assemblies are connected through a plurality of short cylinders 2139. Specifically: at least one mounting hole is provided on each metal bone frame 2133, and the number and distribution positions of the mounting holes are the same as those of the short cylinders 2139 and are aligned vertically one by one.

[0098] As Figure 11 shown in the left figure of , before the marker 21 is separated from the controller 22, the front end of the long snap 223 of the controller 22 extends into the snap socket 211 and rotates to a vertical position, and the long snap 223 is snap-connected to the snap socket 211, so that the marker 21 and the controller 22 are detachably connected. At this time, a limiting rod is installed on the main body bracket 221 of the controller 22. The limiting rod limits the slider 2132 of the elastic clamp 213 of the marker 21 to the front end of the strip-shaped groove, so that the front end of the elastic clamp 213 of the marker 21 is in an open state, forming a jaw. The slider 2132 presses the spring 2138, so that the spring 2138 is in a compressed state.

[0099] As Figure 11 shown in the right figure of , when the jaw of the elastic clamp 213 of the marker 21 passes through the netting, the underwater robot 1 drives the waterproof servo 222 of the controller 22, so that the long snap 223 of the controller 22 rotates back to the horizontal position, the snap socket 211 is disconnected from the long snap 223, and the controller 22 is separated from the marker 21. During the separation process, the limiting rod on the main body bracket 221 of the controller 22 moves away from the slider 2132. The slider 2132 is driven by the restoring force of the spring 2138 and slides to the rear end of the strip-shaped groove, and the jaw closes, so that the elastic clamp 213 of the marker 21 quickly clamps the netting under the restoring force of the spring 2138, and the marker 21 is fixed on the normal netting near the damaged area of the netting.

[0100] Based on the detection of the damaged netting, the present invention fixes the marker on the normal netting near the damaged area of the netting through an elastic clamp, and uses a strobe light to help the diver quickly find the damaged netting, effectively meeting the requirements of damaged netting detection, positioning and maintenance in cage aquaculture.

[0101] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the descriptions in the above embodiments and the specification are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. An underwater net clothing breakage identification method for cage culture, characterized in that: The described method for identifying damaged underwater netting of cage aquaculture includes the following steps: S100) Use the netting damage detection module carried on the underwater robot to perform real-time detection on the netting images collected by the underwater robot, and obtain the detection results; S200) When the underwater robot does not detect damaged netting, proceed to step S500; when the underwater robot detects damaged netting, the underwater robot adjusts its position relative to the damaged netting area and approaches the normal netting near the damaged netting area until the jaws of the elastic clamp of the marker of the netting damage marking device pass through the normal netting near the damaged netting area; S300) The underwater robot controls the controller of the netting damage marking device to fix the marker on the normal netting near the damaged netting area; S400) The underwater robot moves away from the netting to complete the marking of the damaged netting; S500) The underwater robot stops the monitoring operation, or after moving to the next monitoring point, performs the operation according to the same process as steps S100 to S400.

2. The underwater net clothing breakage identification method for cage culture according to claim 1, wherein: In step S100, the netting damage detection module includes a semantic segmentation unit and an image post-processing unit; the semantic segmentation unit is used to segment the netting image using a semantic segmentation model to obtain a corresponding segmentation mask, and several areas of suspected netting damage are marked in the segmentation mask; the image post-processing unit is used to process the segmentation mask to obtain the detection results and output the detection results to the underwater robot; the detection results include no damaged netting and damaged netting. If the detection result is damaged netting, the detection result also includes the position of the damaged netting area.

3. The method for identifying damage to the underwater net of cage culture according to claim 2, characterized in that: The specific process of constructing the semantic segmentation model is as follows: S1) Obtain several netting images in advance, annotate each netting image to obtain the corresponding segmentation mask for each netting image. Each netting image and the corresponding segmentation mask form an image-mask pair, and all image-mask pairs constitute a netting image segmentation dataset; S2) Use the netting image segmentation dataset to train at least one semantic segmentation model to obtain the optimal model weights corresponding to each semantic segmentation model, test each semantic segmentation model respectively, and select the optimal model from all semantic segmentation models as the semantic segmentation model for netting image segmentation according to the test results.

4. The method for identifying damage to the underwater net of cage aquaculture according to claim 2, wherein: The semantic segmentation model adopts the SegFormer model.

5. The method for identifying damage to the underwater net of cage culture according to claim 2, wherein: The process by which the image post-processing unit processes the segmentation mask to obtain the detection results includes the following steps: First, remove the areas of suspected netting damage with a pixel area lower than the preset threshold, and complete each area of suspected netting damage that does not fill the corresponding mesh; Subsequently, detect whether each area of suspected netting damage is damaged: if the area of suspected netting damage is damaged, mark the area of suspected netting damage as a damaged netting area; if the area of suspected netting damage is not damaged, remove the area of suspected netting damage. Finally, if the segmentation mask does not contain a damaged netting area, the detection result of no damaged netting is output; if the segmentation mask contains a damaged netting area, the detection result of the occurrence of damaged netting and the location of the damaged netting area is output.

6. The method for identifying damage to the underwater net of cage aquaculture according to claim 5, characterized in that: The process of detecting whether each to-be-confirmed damaged netting area has damaged netting includes the following steps: Obtain the Euclidean distance between the center point coordinates of each mesh hole in the segmentation mask and the center point coordinates of the to-be-confirmed damaged netting area, and screen out the mesh holes within the preset threshold as the adjacent mesh holes of the to-be-confirmed damaged netting area; Obtain the pixel area of each adjacent mesh hole, screen out the adjacent mesh holes with too large or too small pixel areas, and obtain the average value of the pixel areas of the remaining adjacent mesh holes; Obtain the ratio between the pixel area of the to-be-confirmed damaged netting area and the average value of the pixel areas of the remaining adjacent mesh holes. If the ratio is greater than or equal to 2, the to-be-confirmed damaged netting area has damaged netting. If the ratio is less than 2, the to-be-confirmed damaged netting area has no damaged netting.

7. The method for identifying damage to the underwater net of cage culture according to claim 1, characterized in that: The specific step S300 is as follows: When the jaws of the elastic clamp of the marker of the netting damage marking device pass through the normal netting near the damaged netting area, the underwater robot controls the waterproof servo of the controller of the netting damage marking device to rotate the front end of the long buckle; at this time, the built-in spring of the elastic clamp takes effect and pushes the marker to eject forward; at the same time, the jaws of the elastic clamp quickly clamp under the action of elastic force to fix the marker on the normal netting, and the strobe signal lamp on the marker flashes at a preset certain frequency.

8. A netting breakage marking device applicable to the netting breakage marking method for underwater netting in cage aquaculture as described in any one of claims 1 to 7, characterized in that: The netting damage marking device (2) is installed at the bottom of the underwater robot (1) and includes a marker (21), a controller (22) and a mounting chassis (23); the mounting chassis (23) is fixedly installed on the underwater robot (1), the front end of the mounting chassis (23) is fixedly connected to the rear end of the controller (22), the front end of the controller (22) is detachably connected to the marker (21), and the controller (22) is electrically connected to the underwater robot (1); The marker (21) includes a female buckle base (211), an upper housing (212), an elastic clamp (213), a lower housing (214), and a stroboscopic signal lamp (215); the female buckle base (211) is fixedly installed on the top surface of the upper housing (212), the upper housing (212) and the lower housing (214) are arranged up and down to form a cavity structure, the elastic clamp (213) is arranged in the cavity structure, the jaws of the elastic clamp (213) face forward and the front end extends out of the cavity structure; the stroboscopic signal lamp (215) is installed on the lower housing (214); the controller (22) includes a main body bracket (221), a waterproof servo (222), and a long buckle (223); the waterproof servo (222) is arranged inside the main body bracket (221), the steering wheel of the waterproof servo (222) is fixedly connected to the rear end of the long buckle (223), and the front end of the long buckle (223) is detachably snap-connected to the female buckle base (211); a limiting rod is installed at the front end of the main body bracket (221), and the limiting rod extends into the marker (21) and is arranged in cooperation with the elastic clamp (213).

9. The netting breakage marking device according to claim 8, wherein: The elastic clamp (213) includes a pliers-shaped link mechanism (2131), two elastic support components symmetrically arranged on the upper and lower sides of the pliers-shaped link mechanism (2131), a slider (2132) for realizing the transmission connection between the pliers-shaped link mechanism (2131) and the elastic support components, and a plurality of short cylinders (2139) arranged between the two elastic support components; the pliers-shaped link mechanism (2131) adopts a rhombic four-link mechanism, the ends of the rhombic four-link mechanism are hinged to the slider (2132), and barbs are provided at the front ends; each elastic support component mainly consists of a metal skeleton (2133), a skeleton (2134), a spring block (2137) and a spring (2138); the skeleton (2134) is stacked inside the metal skeleton (2133), a strip-shaped groove is formed in the center of the skeleton (2134), and the slider (2132) is slidably arranged in the strip-shaped groove; the spring block (2137) is installed at the front end of the skeleton (2134), the spring (2138) is arranged between the spring block (2137) and the slider (2132), and the two ends of the spring (2138) respectively abut against the spring block (2137) and the slider (2132); the metal skeletons (2133) in the two elastic support components are connected by a plurality of short cylinders (2139); when the front end of the limiting rod extends into the elastic clamp (213), the front end face of the limiting rod contacts the rear end face of the slider (2132), and the limiting rod can drive the slider (2132) to move forward and approach the spring block (2137), so that the jaws open and compress the spring (2138); when the front end of the limiting rod disengages from the elastic clamp (213), the restoring force of the spring (2138) drives the slider (2132) to move backward and away from the spring block (2137), so that the jaws close.

10. An underwater robot, characterized in that: The underwater robot integrates the net damage detection module according to any one of claims 1 to 7 and the net damage marking device according to any one of claims 8 to 9.

Images