Multi-source data-based sticky fish nest egg collection identification method and system

Abstract

The invention discloses a sticky fish nest egg collection identification method and system based on multi-source data. Underwater camera equipment is arranged in an artificial fish nest area to collect continuous video streams, and a key image sequence reflecting spawning dynamics is extracted from the continuous video streams. And after the image is preprocessed, the improved target detection and identification model is used to carry out positioning and type identification on fish individuals. Meanwhile, the self-adaptive threshold segmentation technology is adopted to detect the roes in the fish nest and estimate the number of the roes. And finally, based on the timestamps and the spatial position information, performing space-time correlation matching on the identified fish individuals and the detected spawning events, establishing a corresponding relation, further analyzing spawning behavior characteristics and nest site utilization efficiency of different fish fingerlings, and generating a visual observation report. According to the invention, the automatic and refined monitoring of the spawning activity of the artificial fish nest is realized, and a scientific basis is provided for the design optimization and resource management of the fish nest.

Description

Technical Field

[0001] This invention relates to the field of ecological monitoring, specifically to a method and system for identifying the egg-gathering nests of sticky fish based on multi-source data. Background Technology

[0002] In assessing the ecological effectiveness of artificial fish nests and protecting fish resources, accurately monitoring fish spawning behavior, identifying spawning species, and counting spawning numbers are crucial core steps. Traditional monitoring methods mainly rely on underwater observation by divers or periodic retrieval of fish nests for manual inspection. These methods are not only labor-intensive and have limited observation frequency, but also significantly interfere with the natural reproductive activities of fish, making it difficult to obtain continuous and objective behavioral data. With the development of underwater camera technology, recording the spawning process via video has become possible, but subsequent analysis still requires extensive manual viewing and interpretation, which is inefficient and highly subjective, failing to meet the needs of large-scale, long-term continuous monitoring.

[0003] In recent years, computer vision technology has begun to be applied to fish identification and counting. Existing solutions mostly focus on monitoring fish schools in open water or static identification of single targets. However, in complex artificial fish nest scenarios, individual fish are often obscured, small in size, and underwater environments present challenges such as uneven lighting, turbid water, and cluttered backgrounds, resulting in a significant decrease in the accuracy of general target detection models. More importantly, most existing technologies treat "fish identification" and "fish egg detection" as two independent tasks, lacking an effective mechanism to accurately correlate specific fish individuals with their spawning behavior and the clusters of eggs they produce. This makes it impossible to answer the crucial ecological question of "which fish laid how many eggs, when, and where."

[0004] Furthermore, there is still a lack of automated, data-driven solutions for the quantitative analysis of spatiotemporal patterns of spawning behavior, and for evaluating the efficiency of different nest structures and placement locations in attracting different fish species to spawn. Therefore, there is an urgent need to develop a comprehensive monitoring method that can operate automatically, continuously, and non-intrusively, deeply integrating fish identification, spawning event detection, and spatiotemporal correlation analysis to ultimately output quantitative evaluation results. This would overcome the limitations of existing technologies and provide reliable technical support for the refined management of artificial fish nests and the protection of fish reproduction. Summary of the Invention

[0005] To address the technical problems mentioned above, this invention provides a method for identifying egg-gathering nests of sticky fish based on multi-source data, comprising the following steps:

[0006] S1. Based on the continuous video stream collected by underwater camera equipment deployed in the artificial fish nest area, obtain the original image sequence reflecting the dynamic process related to spawning;

[0007] S2. Based on the original image sequence, obtain the preprocessed image sequence, and identify and locate individual fish based on the preprocessed image sequence to obtain the identification results including fish species, location and time;

[0008] S3. Based on the preprocessed image sequence, detect fish eggs within the fish nest area and estimate the number of eggs laid;

[0009] S4. Based on the identification results and fish egg detection results, conduct spatiotemporal correlation analysis to establish the correspondence between individual fish and spawning events, and analyze the spawning behavior characteristics based on the correspondence.

[0010] Preferably, S1 includes:

[0011] Based on a continuous video stream, the initial frame is extracted according to a preset time interval;

[0012] Based on the initial frame, the pixel changes between adjacent frames are calculated using the inter-frame difference method, and video segments containing fish activity events are selected based on the area and distribution of the changed regions.

[0013] Based on the selected video segments, key image frames are extracted to form the original image sequence.

[0014] Preferably, in step S2, the method for obtaining the preprocessed image sequence includes:

[0015] Convert the original image sequence to the HSV color space;

[0016] Based on the V channel in the HSV color space, a contrast-limited adaptive histogram equalization process is performed.

[0017] Gaussian filtering is then applied to the image after equalization.

[0018] Based on the filtered image, guided filtering is performed to obtain a preprocessed image sequence.

[0019] Preferably, in step S2, the method for identifying and locating individual fish includes:

[0020] Based on the preprocessed image sequence, the bounding box location information of individual fish is obtained through an object detection network;

[0021] Based on the bounding box location information, individual fish images are cropped from the preprocessed image sequence;

[0022] Fish species are identified using an image classification network based on individual fish images.

[0023] Preferably, S3 includes:

[0024] Based on the fish nest region image in the preprocessed image sequence, a binary image with foreground and background separation is obtained by adaptive threshold segmentation technique;

[0025] Candidate fish egg regions are obtained based on binary images through connected component analysis.

[0026] Based on the morphological characteristics and preset size range of candidate fish egg regions, effective fish egg regions are selected.

[0027] The total number of eggs laid is estimated based on the number and area of ​​effective egg-laying areas.

[0028] Preferably, in step S4, the method for establishing the correspondence between individual fish and spawning events includes:

[0029] Based on the timestamps and location information in the identification results, and the time and location information of the fish egg clusters in the fish egg detection results, a spatiotemporal association window is delineated;

[0030] Search for matching individual fish based on a spatiotemporal correlation window;

[0031] The association confidence level is calculated based on the spatial overlap and temporal proximity between individual fish and fish egg clusters.

[0032] Based on the association confidence level, the correspondence between individual fish and spawning events is determined.

[0033] Preferably, in step S4, after analyzing the spawning behavior characteristics, the method further includes: generating a visualized spawning observation report that includes a list of spawning fish species, an estimated number of spawns, temporal distribution, and spatial density distribution, based on the spawning behavior characteristics and their corresponding relationships.

[0034] The present invention also provides a system for identifying the nests and egg-gathering of sticky fish based on multi-source data. The system is used to implement the method described in any one of claims 1-7, and includes: a data acquisition module, a processing module, a detection module, and an analysis module.

[0035] The acquisition module is an underwater camera device deployed in the artificial fish nest area, used to acquire continuous video streams and obtain raw image sequences reflecting the dynamic process related to spawning;

[0036] The processing module is used to obtain a preprocessed image sequence based on the original image sequence, and to identify and locate individual fish based on the preprocessed image sequence, thereby obtaining an identification result that includes the fish species, location, and time.

[0037] The detection module is used to detect fish eggs within the fish nest area and estimate the number of eggs laid based on the preprocessed image sequence;

[0038] The analysis module is used to perform spatiotemporal correlation analysis based on the identification results and fish egg detection results, establish the correspondence between individual fish and spawning events, and analyze spawning behavior characteristics based on the correspondence.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0040] This invention significantly improves the automation and intelligence of artificial fish nest spawning monitoring, achieving uninterrupted, continuous, and precise observation of the spawning process. By efficiently identifying spawning fish species and accurately locating spawning events, it can automatically generate reliable estimates of spawning numbers and spatiotemporal distribution data, greatly reducing the subjective errors and labor costs of traditional manual observation. The obtained quantitative behavioral characteristics and nest site utilization efficiency index provide objective and detailed data support for scientifically evaluating the effectiveness of fish nest design and optimizing layout schemes, thereby assisting managers in making accurate decisions and effectively improving the protection of fish natural reproduction and resource management. Attached Figure Description

[0041] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] Example 1

[0046] like Figure 1 The diagram shown is a schematic representation of the method flow in this embodiment, and the steps include:

[0047] S1. Based on the continuous video stream collected by underwater camera equipment deployed in the artificial fish nest area, obtain the original image sequence reflecting the dynamic process related to spawning.

[0048] Fixed underwater cameras were deployed in the area where artificial fish nests were constructed to continuously capture video streams of the fish nests and their surrounding waters at a specific frame rate. To extract representative keyframe sequences from the video stream, a method combining time intervals and motion detection was first employed: initial screening was performed at fixed time intervals (e.g., one frame per second), and then the pixel variation amplitude between adjacent frames was calculated using the inter-frame difference method to filter out video segments showing fish entering the fish nests, exhibiting obvious activity, or displaying suspected spawning behavior. Key image frames were extracted from these segments to form an image sequence reflecting the dynamic process related to spawning, which served as the raw image data for subsequent analysis.

[0049] The steps described above using the inter-frame difference method include:

[0050] First, the continuous video stream is decoded into a sequence of grayscale images arranged in chronological order. For each adjacent frame in the sequence, the absolute value of the difference in grayscale values ​​of corresponding pixels is calculated, resulting in a difference image. The brighter areas in this image represent regions where significant changes occur between the two frames, typically corresponding to movement information such as fish swimming, fin movements, or the stirring of water during spawning. By monitoring the total area and spatial distribution of the bright areas in the difference image in real time, when the area of ​​the changed region exceeds a threshold and is concentrated within the fish nest area, it is determined that a meaningful activity event exists near that moment. The relevant frames are then extracted as keyframes and incorporated into the original image data sequence.

[0051] To address issues such as turbidity, uneven lighting, and interference from suspended particles caused by the underwater environment in the original image data, a multi-stage preprocessing process was implemented: First, the image was converted to the HSV color space, and contrast-limited adaptive histogram equalization was applied to the V (luminance) channel to enhance local details and suppress global illumination fluctuations. Then, Gaussian filtering was used to smooth the image, reducing noise particle interference while maintaining the edge integrity of fish bodies and eggs. To further improve the separation between the target and the background, guided filtering was used to perform edge-preserving filtering, enhancing texture information while removing noise. After these processes, a preprocessed image with significantly improved clarity and contrast was obtained, providing high-quality input for subsequent fish identification and egg detection.

[0052] S2. Based on the original image sequence, obtain the preprocessed image sequence, and identify and locate individual fish based on the preprocessed image sequence to obtain the identification results including fish species, location and time.

[0053] After obtaining the preprocessed image sequence, the first step is to locate and extract features from individual fish. To efficiently and accurately locate fish targets within nests in complex underwater environments, this embodiment employs an improved version of the single-stage object detection algorithm YOLOv5 for initial detection. This algorithm treats object detection as a unified regression problem, directly predicting bounding box coordinates and class probabilities from image pixels within a single neural network. Its core lies in dividing the input image into an S×S grid, with each grid cell responsible for predicting the object falling into its center. Each prediction includes bounding box information (center coordinates x, y, width w, height h), a confidence score, and a class probability.

[0054] In this embodiment, an improved YOLOv5 target detection algorithm is used for preliminary detection. While retaining the original efficient single-stage detection framework, its network structure has been specifically optimized to address the characteristics of small underwater fish targets, numerous environmental disturbances, and uneven lighting. The entire network can be divided into four core parts: Backbone, Neck, Head, and an introduced dedicated module.

[0055] (1) Backbone network

[0056] The backbone network is responsible for extracting multi-level features from the input image. The improved version retains the CSPDarknet53 architecture but has undergone lightweight adjustments. Its core is the combination of the Cross-Stage Partial Network (CSPNet) and Darknet convolutional blocks. Specifically, the input image first undergoes a focus module for slicing, reorganizing spatial information into the channel dimension to achieve downsampling without information loss. It then passes through multiple C3 modules (CSP Bottleneck with 3 convolutions). Each C3 module divides the input feature map into two parts along the channel. One part undergoes feature extraction through multiple Bottleneck residual blocks consisting of 1x1 and 3x3 convolutions, while the other part is directly short-connected. Finally, the two parts are concatenated. This structure deepens the network while mitigating gradient vanishing and maintaining computational efficiency.

[0057] To enhance the network's ability to perceive small underwater targets and texture features, we embedded a Spatial and Channel Attention Fusion Module (SCAF) into the deep C3 module. This module computes channel attention and spatial attention in parallel.

[0058] (2) Neck network

[0059] The neck network employs an improved FPN+PANet (Path Aggregation Network) structure, but replaces some standard convolutional blocks in the traditional upsampling-concatenation-convolution path with the more efficient C3SAC module (C3 with SpatialAttention and CoordConv). Building upon C3, this module introduces a coordinate convolution (CoordConv) before the 3x3 convolution in the Bottleneck area, concatenating the normalized coordinates of pixels as two additional channels with the input feature map. This enhances the model's perception of spatial location information, which is crucial for locating scattered eggs and fish. Simultaneously, spatial attention is preserved in the feature fusion path, making the fusion process more focused on regions with semantic information.

[0060] (3) Detection head

[0061] The detection head adopts the decoupled head design of YOLOv5, which separates the classification and bounding box regression tasks, thus helping to improve accuracy. For each feature map scale, the detection head predicts the class probability and bounding box information (coordinates, size, confidence score) through two parallel convolutional branches.

[0062] (4) Core formula and loss function

[0063] The output of the detection head is mapped onto three different scale grids to detect targets of varying sizes. Each grid cell predicts multiple bounding boxes. The bounding box prediction uses the following formula:

[0064]

[0065] in, It is the offset predicted by the network. It's the Sigmoid function, which restricts the offset to between 0 and 1. It is the coordinate of the top left corner of the grid cell. These are the preset width and height of the anchor frame. The improved version regenerates the anchor frame size on a specific underwater fish dataset using K-means++ clustering to adapt to the actual distribution of underwater targets.

[0066] The model's total loss function is as follows:

[0067]

[0068] in, It is the bounding box regression loss. It is the target confidence loss. It is classification loss. It is the weighting coefficient.

[0069] After locating individual fish, they are cropped from the original image and their size is normalized before being used as input to the subsequent species identification model. The convolutional neural network model used for species identification is a specially designed and trained deep architecture. This model is based on a ResNet-50 backbone, utilizing its residual learning units to mitigate the vanishing gradient problem in deep networks, and is adaptively tuned for underwater fish identification tasks. The input is a normalized RGB image of a single fish (e.g., 224x224 pixels). The initial part of the model is a 7x7 convolutional layer with a stride of 2, outputting a feature map with 64 channels, followed by a 3x3 max-pooling layer with a stride of 2 for fast downsampling and low-level feature extraction. This is followed by four consecutive convolutional blocks (Conv2_x to Conv5_x) with a bottleneck design. Each bottleneck block contains three convolutional layers: a 1x1 convolution for dimensionality reduction, a 3x3 convolution for spatial feature extraction, and another 1x1 convolution for dimensionality recovery and addition. Each convolutional layer is followed by batch normalization and the ReLU activation function. Specifically, for an input feature map X, the learning process of a residual block can be represented as:

[0070]

[0071] in, and For convolution operations, It is the ReLU activation function. For batch normalization.

[0072] The block output is:

[0073]

[0074] in, It is a residual connection implemented through 1x1 convolution when the input and output dimensions do not match.

[0075] Building upon the original ResNet-50, this embodiment removes the final downsampling layer before the global average pooling layer after Conv5_x to preserve higher-resolution feature maps, which is crucial for distinguishing texture details (such as spots and stripes). Subsequently, a custom attention module is integrated. This module first re-labels the importance of each channel of the feature map through a channel attention submodule. The specific process is as follows:

[0076] For the input feature map (H and W represent the height and width of the feature, respectively). Global average pooling and global max pooling are performed to obtain two C-dimensional channel descriptors. These two descriptors are then input into a shared multilayer perceptron (MLP, consisting of one hidden layer). The features output by the MLP are element-wise summed and then passed through a sigmoid activation function to obtain the channel attention weight vector. Final output features ,in This represents channel-by-channel multiplication. Next, the spatial attention submodule... Average pooling and max pooling are performed along the channel dimension, and the two results are concatenated into a 2-channel feature map. This map is then passed through a 7x7 convolutional layer and a sigmoid activation function to generate a spatial attention weight map. Final output features The introduced attention mechanism enables the model to adaptively focus on the most discriminative areas of the fish's body surface (such as specific markings and fin shapes) while suppressing interference from murky backgrounds or shadows.

[0077] The feature maps enhanced by the attention module are finally passed through a global average pooling layer, which compresses the two-dimensional feature map of each channel into a scalar, forming a fixed-length feature vector. This vector is then fed into a fully connected layer with the number of neurons equal to the number of fish species to be identified (e.g., carp, crucian carp, catfish, etc.). Following the fully connected layer is a softmax activation function, which transforms the output into a probability distribution for each fish species. Given a feature vector... The weight matrix of the fully connected layer is , bias is Then the unnormalized score for category j The Softmax function calculates the probability of class j as follows:

[0078]

[0079] in, This represents the total number of categories. The network is trained end-to-end using a large-scale labeled underwater fish image dataset, with the loss function being cross-entropy loss.

[0080]

[0081] in, It is a category The unique hot-coded real label.

[0082] S3. Based on the preprocessed image sequence, detect fish eggs in the fish nest area and estimate the number of eggs laid.

[0083] For images of fish nest areas, an adaptive threshold segmentation technique is used to separate the foreground (fish eggs and possible impurities) from the background. Based on the morphological characteristics (such as approximate circularity and size range) and color features of the fish eggs, the segmented regions are screened to initially identify candidate fish egg regions. Connected component analysis is used to merge adjacent similar pixel regions, and parameters such as the area and circularity of each connected component are calculated. Combined with a preset range of fish egg sizes, it is determined whether a region is a valid fish egg region. The number of all valid fish egg regions is counted, and the total number of eggs is estimated based on the ratio of their average area to the known actual size of the fish eggs.

[0084] The adaptive thresholding method fully considers the spatial variations in grayscale values ​​caused by uneven illumination, local turbidity differences, and shadows on the surface of fish nests in underwater images. Its core principle is that the binarization threshold of each pixel in the image is not fixed, but dynamically calculated based on the grayscale characteristics of the local neighborhood around that pixel.

[0085] First, input a grayscale image that has been denoised using a Gaussian filter. For each pixel in the image... A local window of size N×N (e.g., 31x31 pixels) is defined centered on the given element. The weighted average of the grayscale values ​​of all pixels within this window is calculated as the local threshold. Commonly used calculation methods include:

[0086] (1) The average gray value of the local neighborhood;

[0087] (2) Gaussian weighted average of the local neighborhood, which gives higher weight to the central region and produces a smoother threshold surface. To enhance the robustness of segmentation, a constant offset C (e.g., -5 to -15) is usually introduced. The final threshold calculation formula is:

[0088]

[0089] in, This is either a local mean or a weighted mean. Then, the grayscale value of the current pixel is... With dynamic threshold Comparison: If If a pixel is selected, it is set as the background (e.g., 0, black); otherwise, it is set as the foreground (e.g., 255, white). After traversing all pixels, a binarized segmented image is obtained.

[0090] In underwater fish nest scenes, foreground objects (fish eggs) typically appear as slightly darker or brighter areas than their surrounding background in localized regions due to their translucency and contrast with the attached substrate. Adaptive thresholding effectively addresses the gradual shadows created by curved surfaces at the edges of the fish nest, as well as localized brightness fluctuations caused by suspended matter in the water. Through this dynamic adjustment, even under uneven overall lighting, potential fish egg regions can be relatively consistently segmented from the background while suppressing most interference from non-uniform backgrounds.

[0091] S4. Based on the identification results and fish egg detection results, conduct spatiotemporal correlation analysis to establish the correspondence between individual fish and spawning events, and analyze the spawning behavior characteristics based on the correspondence.

[0092] The individual fish identified in S2 are precisely matched with the spawning events detected in S3. Specifically, based on timestamps and spatial location information, a correlation analysis is performed on fish appearing in the same nest area during the same time period and the clusters of fish eggs appearing around them to establish a correspondence between "fish - spawning behavior - number of fish eggs".

[0093] Specifically, spatiotemporal correlation analysis is performed based on the results output by the fish detection and recognition model in S2—including the bounding box coordinates, species labels, and corresponding timestamps of the fish identified in each frame. In this embodiment, a spatiotemporal correlation window is set: in the time dimension, a preset time interval (e.g., 60 seconds) is traced back from the moment the fish egg cluster is first detected; in the spatial dimension, an extended spatial region (e.g., the bounding box is expanded by 20%) is defined based on the geometric center of the fish egg cluster. The system searches for all identified and located fish individuals within this spatiotemporal window. Candidate association pairs are established by calculating the spatial overlap (e.g., IoU index) between the fish bounding box and the fish egg cluster region, as well as the temporal proximity of their occurrence. To accurately establish the correspondence between "fish-spawning behavior-number of fish eggs," the system uses a weighted scoring model based on multidimensional features (including spatial distance, time interval, typical duration of spawning behavior of fish species, etc.) to calculate a confidence score for each candidate association pair and select the match with the highest score, thereby attributing a specific fish egg cluster event to a specific fish individual or its transient combination.

[0094] Building upon this foundation, we further analyzed the spawning preferences of different fish species in artificial nests, including behavioral characteristics such as visit frequency, spawning duration, single spawning quantity, and nest location selection. We then used a statistical model to quantify and evaluate the utilization efficiency of each fish species for specific nest structures or locations. The core of this model is the definition of a "nest location utilization efficiency index." This index comprehensively considers multiple observed variables: for a fish species i and a specific nest j, the variables include the fish species' visit frequency in nest j (total occurrences), the proportion of effective spawning events (the proportion of visits associated with spawning), the average single spawning quantity (the mean number of eggs in associated events), and the spawning location retention rate (the proportion of eggs not preyed upon or scattered based on subsequent frame analysis). Using principal component analysis or expert weighting, we assigned weights to each variable and calculated the standardized utilization efficiency index for each "fish species-nest" combination. By comparing the indices of fish nests of the same fish species in different structures or locations, the quality of fish nest design parameters can be evaluated. By comparing the indices of different fish species within the same fish nest, the diversity of fish species attracted by the nest and the dominant fish species can be analyzed. These quantitative results provide objective and detailed data for targeted optimization of fish nests (such as adjusting the opening size, internal complexity, placement depth, and flow direction).

[0095] Finally, a visualized spawning observation report is generated, including a list of spawning fish species, estimated spawning numbers for each species, temporal distribution curves, and a geographic information system-based heatmap of spawning density, visually displaying the spatiotemporal pattern of spawning activity within the fish nests. This report provides data-driven decision support for managers, enabling them to evaluate the design effectiveness of existing fish nests, optimize nest materials, structures, and layout schemes, and offer scientific recommendations for nest maintenance, replenishment, or adjustments during the breeding season. Ultimately, this achieves dynamic and refined management of artificial fish nest resources, enhancing the effectiveness of natural fish reproduction conservation.

[0096] Example 2

[0097] This embodiment also provides a system for identifying the egg-laying behavior of sticky fish nests based on multi-source data. The system includes an acquisition module, a processing module, a detection module, and an analysis module. The acquisition module is an underwater camera deployed in the artificial fish nest area to acquire continuous video streams and obtain raw image sequences reflecting the dynamic process related to spawning. The processing module is used to obtain preprocessed image sequences based on the raw image sequences, and to identify and locate individual fish based on the preprocessed image sequences, obtaining identification results including fish species, location, and time. The detection module is used to detect fish eggs in the nest area and estimate the number of eggs laid based on the preprocessed image sequences. The analysis module is used to perform spatiotemporal correlation analysis based on the identification results and fish egg detection results, establish the correspondence between individual fish and spawning events, and analyze spawning behavior characteristics based on the correspondence.

[0098] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for identifying egg-gathering nests of sticky fish based on multi-source data, characterized in that, Includes the following steps: S1. Based on the continuous video stream collected by underwater camera equipment deployed in the artificial fish nest area, obtain the original image sequence reflecting the dynamic process related to spawning; S2. Based on the original image sequence, obtain the preprocessed image sequence, and identify and locate individual fish based on the preprocessed image sequence to obtain the identification results including fish species, location and time; S3. Based on the preprocessed image sequence, detect fish eggs within the fish nest area and estimate the number of eggs laid; S4. Based on the identification results and fish egg detection results, conduct spatiotemporal correlation analysis to establish the correspondence between individual fish and spawning events, and analyze the spawning behavior characteristics based on the correspondence.

2. The method for identifying the nest-egg collection of sticky fish based on multi-source data according to claim 1, characterized in that, S1 includes: Based on a continuous video stream, the initial frame is extracted according to a preset time interval; Based on the initial frame, the pixel changes between adjacent frames are calculated using the inter-frame difference method, and video segments containing fish activity events are selected based on the area and distribution of the changed regions. Based on the selected video segments, key image frames are extracted to form the original image sequence.

3. The method for identifying the nest-egg collection of sticky fish based on multi-source data according to claim 1, characterized in that, In step S2, the method for obtaining the preprocessed image sequence includes: Convert the original image sequence to the HSV color space; Based on the V channel in the HSV color space, a contrast-limited adaptive histogram equalization process is performed. Gaussian filtering is then applied to the image after equalization. Based on the filtered image, guided filtering is performed to obtain a preprocessed image sequence.

4. The method for identifying the nest-egg collection of sticky fish based on multi-source data according to claim 1, characterized in that, In step S2, the method for identifying and locating individual fish includes: Based on the preprocessed image sequence, the bounding box location information of individual fish is obtained through an object detection network; Based on the bounding box location information, individual fish images are cropped from the preprocessed image sequence; Fish species are identified using an image classification network based on individual fish images.

5. The method for identifying the egg-gathering nests of sticky fish based on multi-source data according to claim 1, characterized in that, S3 includes: Based on the fish nest region image in the preprocessed image sequence, a binary image with foreground and background separation is obtained by adaptive threshold segmentation technique; Candidate fish egg regions are obtained based on binary images through connected component analysis. Based on the morphological characteristics and preset size range of candidate fish egg regions, effective fish egg regions are selected. The total number of eggs laid is estimated based on the number and area of ​​effective egg-laying areas.

6. The method for identifying the egg-gathering nests of sticky fish based on multi-source data according to claim 1, characterized in that, In S4, the method for establishing the correspondence between individual fish and spawning events includes: Based on the timestamps and location information in the identification results, and the time and location information of the fish egg clusters in the fish egg detection results, a spatiotemporal association window is delineated; Search for matching individual fish based on a spatiotemporal correlation window; The association confidence level is calculated based on the spatial overlap and temporal proximity between individual fish and fish egg clusters. Based on the association confidence level, the correspondence between individual fish and spawning events is determined.

7. The method for identifying the nest-egg collection of sticky fish based on multi-source data according to claim 1, characterized in that, In step S4, after analyzing the spawning behavior characteristics, the method further includes: generating a visualized spawning observation report that includes a list of spawning fish species, an estimate of the number of eggs laid, and the temporal and spatial density distributions, based on the spawning behavior characteristics and their corresponding relationships.

8. A system for identifying the nests and egg-gathering of sticky fish based on multi-source data, the system being used to implement the method described in any one of claims 1-7, characterized in that, include: Acquisition module, processing module, detection module, and analysis module The acquisition module is an underwater camera device deployed in the artificial fish nest area, used to acquire continuous video streams and obtain raw image sequences reflecting the dynamic process related to spawning; The processing module is used to obtain a preprocessed image sequence based on the original image sequence, and to identify and locate individual fish based on the preprocessed image sequence, thereby obtaining an identification result that includes the fish species, location, and time. The detection module is used to detect fish eggs within the fish nest area and estimate the number of eggs laid based on the preprocessed image sequence; The analysis module is used to perform spatiotemporal correlation analysis based on the identification results and fish egg detection results, establish the correspondence between individual fish and spawning events, and analyze spawning behavior characteristics based on the correspondence.

Images