A method and system for evaluating relative abundance of fish in field wetland based on fixed underwater video
By continuously acquiring video data using fixed underwater cameras and combining it with a target detection model, a list of fish species and a basic image database for the monitoring area were established. The target detection model was trained, and various time windows were used. With the help of an expert detection model, the problem of fish relative abundance assessment models in existing technologies was solved. This resulted in a low-interference and standardized technical solution, addressing the issues of empirically determined sampling duration and lack of consideration for species differences in existing technologies. This enabled a scientific and repeatable assessment of fish relative abundance.
Patent Information
- Application Number
- CN202611306393.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-11-03
AI Technical Summary
In existing fixed underwater video monitoring methods, the sampling duration mainly relies on empirical settings and lacks a unified basis for comparison, making it difficult for MaxN results to accurately characterize the relative abundance of fish, and the differences in activity characteristics of different species are not taken into account.
By continuously acquiring video using fixed underwater cameras, a list of fish species and a basic image database for the monitoring area are established. A target detection model is trained, and a MaxN calculation is constructed using a long window of multiple sampling times. The recommended sampling duration is determined by a rank-based nonparametric multiple comparison method, a relative abundance assessment window is constructed, and the relative abundance assessment results are output.
It achieves low-interference, standardized fish relative abundance assessment, reduces manual processing workload, improves the traceability and comparability of monitoring data, and ensures the scientific nature and repeatability of assessment results.
Smart Images
Figure CN123033103A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of aquatic organism video monitoring, computer vision and ecological monitoring technology, specifically a method and system for assessing the relative abundance of fish in wild wetlands based on fixed underwater video. Background Technology
[0002] With the development of underwater photography equipment, data storage, and computer vision technology, fixed underwater video is gradually being applied to fish monitoring in wetlands. This method can continuously record fish activities within a fixed camera field of view with minimal disturbance to the natural behavior of fish, and the resulting videos can be repeatedly verified. By standardizing equipment deployment, shooting parameters, and data processing procedures, the comparability and traceability of monitoring data from the same site at different times, as well as monitoring data from different independently calibrated sites, can be improved.
[0003] By using a target detection model to identify, classify, and count underwater video images, a frame-by-frame species count record can be generated, including image timestamps, species categories, and the number of detections. For continuous video under a fixed camera field of view, the same individual may appear continuously in multiple consecutive frames or re-enter the camera field of view after leaving it; directly accumulating the detection counts in different image frames may result in duplicate counting. Therefore, this invention uses MaxN to represent the relative abundance of fish under a fixed camera field of view. MaxN refers to the maximum number of individuals of the same species appearing simultaneously in any single frame image within a preset sampling window, used to represent the relative abundance within the corresponding sampling window, and does not represent the absolute number of individuals in the monitoring area.
[0004] Existing fixed underwater video surveys often rely on empirically determined sampling durations, lacking a unified method for comparing the empirical distribution of MaxN obtained from different sampling durations on the same data basis, determining an appropriate sampling duration, and conducting relative abundance assessments accordingly. When the sampling duration is too short, fish have fewer opportunities to enter the camera's field of view, and the obtained MaxN may be insufficient to characterize the relative abundance of the corresponding time period. Conversely, when the sampling duration is too long, it increases the workload of video storage, model inference, and result verification. Furthermore, different species exhibit varying activity characteristics and their probability of detection by camera equipment, making it difficult to consider different species using a single empirical sampling duration. Therefore, there is an urgent need to establish a method and system capable of quantifying the MaxN response to sampling duration, determining recommended sampling durations for species, and assessing relative abundance according to these recommended sampling durations. Summary of the Invention
[0005] To address the issues that fixed underwater video sampling durations rely primarily on experience, lack a unified basis for comparing MaxN values obtained from different sampling durations, and are difficult to assess relative abundance of fish under standardized sampling durations, this invention provides a method and system for assessing the relative abundance of fish in wild wetlands based on fixed underwater video.
[0006] To achieve the above objectives, the present invention provides the following technical solution, which specifically includes the following steps: S1. Fixed underwater video acquisition and metadata recording. Fixed underwater camera equipment is deployed in the wetlands to ensure that the equipment maintains a stable installation position, shooting direction, and field of view during the monitoring period; continuously acquire videos of fish activity, and record the site identification, geographical location, equipment identification, video start and end time, time zone, video resolution, equipment maintenance time, and any abnormal situations.
[0007] Fixed underwater camera equipment should maintain a stable installation position, shooting direction, and field of view during the monitoring period; select an installation location that avoids continuous obstruction of the target monitoring area by aquatic plants, floating objects, sediments, or other obstacles; inspect and clean the camera lens or transparent protective cover according to the preset maintenance cycle, and remove mud, algae, and other attachments from its surface to reduce field of view obstruction and image quality degradation caused by lens contamination; at the same time, record the site identification, geographical location, equipment identification, video start and end time, time zone, video resolution, equipment maintenance time, and abnormal situation records.
[0008] S2. Construction of the List of Monitored Species and Basic Image Database. Based on historical survey data and field pre-survey results, identify the potential fish species distributed within the monitoring area, and determine the dominant species, common species, and key monitoring species. Collect fish images covering different lighting conditions, water turbidity, individual posture, individual size, and degree of occlusion for the target monitoring species, and establish a basic image database of fish in the monitoring area.
[0009] S3. Object Detection Model Training and Validation. Fish target bounding boxes and species categories in the basic fish image database are manually labeled. The labeled images are divided into independent training, validation, and test sets, using at least one of the following as grouping units: video clips, sampling dates, or monitoring stations. The object detection model is trained using the training set, and the model parameters are optimized and confidence thresholds for each species are determined using the validation set. Precision, recall, F1 score, mean precision, or mean mean precision are calculated using the test set to evaluate the object detection model's recognition performance.
[0010] S4. Frame-by-Frame Species Identification and Counting. Images are extracted from fish activity videos at preset time intervals. A trained target detection model is used to identify, classify, and count the images, forming a frame-by-frame species count record containing image identifiers, image timestamps, species categories, confidence levels, and detection counts. The date is then recorded. The timestamp of the i-th frame image is . The number of species s detected in the i-th frame image is denoted as . .
[0011] S5. Long-term window construction and MaxN calculation during multi-sampling. Two or more candidate sampling durations are set, and these candidate sampling durations constitute a set: For date d and candidate sampling duration Based on the base time of that date Starting from the sliding step size Construct the k-th time window: Candidate sampling duration within the corresponding date The corresponding number of time windows is: In the formula, m represents the number of candidate sampling durations; Indicates the sampling duration of the j-th candidate; and These represent the start and end times of monitoring on date d, respectively. Indicates the candidate sampling duration The corresponding sliding step size, and 0 < ≤ ; Indicates the date d during the candidate sampling duration. The k-th time window below; Indicates the total number of corresponding time windows; when = When 0 < 0, adjacent time windows do not overlap; when 0 < 0, adjacent time windows do not overlap. < At that time, adjacent time windows partially overlap.
[0012] The species count records for each frame are assigned to the corresponding time windows based on the image timestamps, and MaxN is calculated according to the following formula: In the formula, Indicates species s within a time window The maximum number of individuals that can appear simultaneously in any single frame of an image; when species s is not detected in the corresponding image, Record it as 0.
[0013] Nonparametric multiple comparisons of the MaxN distribution. MaxN values obtained for each species under different candidate sampling durations within a preset monitoring period are summarized to form MaxN sample sets corresponding to each candidate sampling duration. For the same species, a rank-based nonparametric multiple comparison method is used, with Tukey-type pairwise comparisons to compare the MaxN distributions corresponding to different candidate sampling durations. The relative contrast effect estimate, confidence interval, test statistic, and p-value corrected for multiple comparisons are obtained. When the corrected p-value is less than a preset significance level, a significant difference is determined between the MaxN distributions corresponding to the two candidate sampling durations. When the corrected p-value is not less than the preset significance level, no significant distributional difference is detected under the statistical test rules.
[0014] S6. Determine the recommended sampling duration. Place species s within the candidate sampling duration... and The corrected P-value between them is denoted as Let the preset significance level be denoted as α, and determine the recommended sampling duration according to the following formula: When there are no candidate sampling durations that satisfy the above conditions, let: When the corrected P-value between a candidate sampling duration and all longer candidate sampling durations is not less than α, it is determined that no significant distribution difference was detected under the preset statistical test rules, and the MaxN empirical distribution of the corresponding species tends to stabilize as the sampling duration increases; the shortest candidate sampling duration that meets the above conditions is determined as the recommended sampling duration for the species.
[0015] S7. Relative Abundance Assessment and Results Output. Sampling duration is determined according to the recommended duration for species s. Construct a relative abundance assessment window and determine the relative abundance index for the corresponding species according to the following formula: In the formula, This represents the relative abundance index of species *s* within the k-th relative abundance assessment window on date *d*, used to compare relative abundance changes between different monitoring dates or periods under conditions of fixed camera equipment, fixed camera field of view, and consistent video acquisition. It outputs the MaxN result, MaxN distribution, relative contrast effect estimate, confidence interval, test statistic, p-value corrected for multiple comparisons, recommended sampling duration, and relative abundance assessment results for each species. It also saves the target detection model version, species confidence threshold, candidate sampling duration, sliding step size, preset significance level, and nonparametric multiple comparison parameters.
[0016] MaxN is used to characterize the changes in the relative abundance of species under fixed camera field of view and consistent video acquisition conditions, and is not used to represent the absolute population size of species in the monitoring area. MaxN results obtained from different monitoring stations, camera equipment, shooting directions or field of view should be compared after maintaining consistency under the corresponding acquisition conditions or after independent calibration.
[0017] The present invention also provides a system for assessing the relative abundance of fish in wild wetlands based on fixed underwater video, including a fixed underwater video acquisition device and a data processing device; The data processing device includes a frame-by-frame species identification and counting module, a candidate sampling duration setting module, a time window construction module, a MaxN calculation module, a MaxN empirical distribution construction and comparison module, a recommended sampling duration determination module, and a relative abundance evaluation module. The time window construction module is used to group the frame-by-frame species counting records by time according to the candidate sampling duration and the sliding step size. The MaxN calculation module is used to calculate the maximum number of individuals of each species that appear simultaneously in any single frame image within each time window. The relative abundance assessment module is used to construct a relative abundance assessment window according to the recommended sampling duration for each species, and the MaxN within the corresponding window represents the relative abundance of each species.
[0018] Compared with the prior art, the present invention has at least the following beneficial effects: (1) The present invention uses a fixed underwater camera to continuously collect videos of fish activities, which can carry out fish monitoring in wild wetlands in a low-interference, non-contact and non-invasive manner, and improve the traceability of monitoring data through complete metadata recording; (2) By establishing a list of fish species and a basic image database in the monitoring area, this invention trains a target detection model to identify, classify and count fish in video images, which can reduce the amount of manual processing and improve the consistency of the fish identification and counting process; (3) The present invention defines the time grouping method under different candidate sampling durations by uniformly defining the window length and sliding step size, and calculates MaxN on the same frame-by-frame species counting record, so that the comparison of different sampling durations has a consistent data basis.
[0019] (4) This invention compares the MaxN distributions corresponding to different candidate sampling durations by using rank-based nonparametric multiple comparisons and Tukey-type pairwise comparisons, and determines the shortest candidate sampling duration that meets the preset statistical test rules, providing a quantitative basis for the selection of sampling duration in underwater video surveys; (5) The present invention constructs a relative abundance assessment window according to the recommended sampling duration of each species, and uses MaxN in the corresponding window to characterize the relative abundance of species, which can reduce the impact of sampling duration differences on the relative abundance assessment results; at the same time, it saves the model, window and statistical test parameters, and improves the repeatability of the assessment process. (6) This invention establishes a standardized frame-by-frame species counting record through the synergy of fixed underwater video acquisition and frame-by-frame identification of target detection models, providing a consistent data foundation for subsequent analysis; through the synergy of multi-sampling duration window construction and MaxN calculation, the comparison of different sampling durations has a unified data basis, transforming duration selection from empirical judgment into a quantifiable problem; through the synergy of rank-based Tukey-type nonparametric multiple comparisons and species-specific duration recommendations, the shortest sampling duration that meets the statistical stability condition is determined for fish with different activity characteristics; through the synergy of re-evaluating relative abundance according to the recommended duration for each species, it ensures that each species produces standardized indicators under the appropriate duration, bridging the incomparability caused by differences in sampling duration; through the synergy of the above technical solutions, this invention achieves the overall technical effect of adaptively determining sampling duration and standardizing the evaluation of fish relative abundance in a low-interference manner, effectively solving the problems of empirical sampling duration, neglect of species differences, and incomparability of evaluation results, providing a scientific, efficient, and repeatable technical solution for long-term monitoring and ecological management of fish in wild wetlands. Attached Figure Description
[0020] Figure 1 This is a flowchart illustrating the steps of a method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video, as described in this invention. Figure 2 This is a diagram of the fixed underwater video acquisition device of the present invention; Figure 3 This is a schematic diagram of the fixed underwater video acquisition device, system composition, and field deployment of the present invention. Figure 3 The image in section 'a' shows the underwater video acquisition device. Figure 3 Figure b is a schematic diagram of the actual field deployment of the underwater video device. Figure 3 The diagram in C is a schematic of a solar-powered field deployment. Figure 3 The diagram below, d, shows a fixed underwater video acquisition device.
[0021] Figure 4 This is an empirical distribution map of MaxN for 10 target fish species under six candidate sampling durations of 30 minutes, 60 minutes, 180 minutes, 360 minutes, 720 minutes, and 1440 minutes, as shown in this embodiment of the invention. Detailed Implementation
[0022] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are used to illustrate the implementation process of the technical solution of the present invention and do not constitute a limitation on the model of fixed underwater camera equipment, the type of target detection model, the number of target fish species, the candidate sampling time, the sliding step size, or the statistical test parameters.
[0023] like Figure 1 As shown, a method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video includes the following steps: S1. Fixed underwater video acquisition and metadata recording: By deploying fixed underwater camera equipment, continuously acquire videos of fish activities and record site information, equipment information and time information; like Figure 3 As shown, this embodiment uses a fixed underwater video acquisition device to continuously record fish activities in the target waters.
[0024] like Figure 2 As shown, the fixed underwater video acquisition device includes an underwater camera unit 1, a solar power supply unit 2, a control and power supply unit 3, a data transmission unit 4, a video storage unit 5, and a wireless communication unit 6. The device's operating status and video footage can be viewed through a remote monitoring terminal.
[0025] The underwater camera unit 1 is fixedly deployed in the target monitoring water area, ensuring that its installation position, shooting direction, and field of view remain stable throughout the monitoring period. The installation location is chosen to avoid continuous obstruction of the target monitoring area by aquatic plants, floating objects, sediment, or other obstacles. In this embodiment, the fixed underwater video acquisition device is inspected and the camera lens or transparent protective cover is cleaned every 3 days. In other embodiments, the maintenance cycle can be adjusted according to changes in water turbidity, algae attachment rate, sediment accumulation, and image clarity.
[0026] During video acquisition, metadata such as site identifier, geographical location, device identifier, video start and end time, time zone, video resolution, image extraction interval, device maintenance time, and abnormal conditions are saved, so that video files, image files, and subsequent analysis results can correspond to each other.
[0027] The monitoring period for this embodiment was from May 4, 2025 to June 4, 2025, a total of 32 consecutive days, resulting in 541,733 frame-by-frame records with image timestamps. The target monitored species included 10 fish species: crucian carp, common carp, catfish, topmouth gudgeon, plateau loach, *Goblinus borscheriensis*, *Goblinus lazarovichthys*, loach, *Goblinus simonii*, and *Goblinus sinensis*. The dates, number of records, and number of species mentioned above are used to illustrate the data scale of this embodiment.
[0028] S2. Construct a list of monitored species and a basic image database of fish; S3. Manually label the bounding boxes and species categories of fish targets in the basic fish image database, and divide the labeled images into independent training, validation and test sets according to a preset ratio; train the target detection model to obtain the trained target detection model. A fish species list for the monitoring area was compiled based on historical survey data and on-site preliminary survey results. Fish images covering different lighting conditions, water turbidity, individual postures, sizes, and degrees of occlusion were collected to establish a basic fish image database. Fish target bounding boxes and species categories were manually labeled. Using at least one of the following as grouping units—video clips, sampling dates, or monitoring sites—the labeled images were divided into independent training, validation, and test sets, ensuring that images within the same group were only assigned to one dataset.
[0029] This embodiment employs the YOLO26 target detection model. The YOLO26 target detection model is trained using a training set, and its parameters are optimized and confidence thresholds for each target species are determined using a validation set. Precision and recall are calculated using an independent test set to evaluate the fish recognition performance of the target detection model. The validated YOLO26 target detection model is used for fish identification, species classification, and individual counting in images extracted from underwater videos.
[0030] S4. Based on the acquired continuous video and the target detection model, extract image frames at preset fixed time intervals, run the target detection model on each image frame, output the fish target bounding box, species, confidence score and detection count, generate a frame-by-frame species count record, and output a structured frame-by-frame species count record. S5. Based on the frame-by-frame species counting record, construct the corresponding time window according to the preset candidate sampling duration, calculate the maximum number of individuals MaxN of each species that appear simultaneously in any single frame image within each time window, and obtain the MaxN value sequence of each species in each time window under each candidate sampling duration. Images are extracted from fish activity videos at preset time intervals; in this embodiment, the main image extraction interval is approximately 5 seconds. A trained target detection model is used to identify, classify, and count each image, forming a frame-by-frame species count record that includes image identifier, image timestamp, species category, confidence level, and detection count. When no species s is detected in an image, the detection count of species s in that image is recorded as 0.
[0031] The candidate sampling duration set in this embodiment is: T = {30 minutes, 60 minutes, 180 minutes, 360 minutes, 720 minutes, 1440 minutes}.
[0032] In this embodiment, h_j = T_j, meaning the sliding step size corresponding to each candidate sampling duration is equal to the corresponding window length. Adjacent and non-overlapping time windows are divided starting from 00:00:00 on each date. The daily window numbers corresponding to the six candidate sampling durations are 48, 24, 8, 4, 2, and 1, respectively. The total number of windows for 32 consecutive days are 1,536, 768, 256, 128, 64, and 32, respectively.
[0033] For species s, date d, the j-th candidate sampling duration T_j, and the k-th time window, the maximum number of species s detected in each frame of the corresponding time window is determined as MaxN, i.e., MaxN(s,d,k,j)=max{c(s,d,i)|τ(d,i)∈W(d,k,j)}. MaxN represents the maximum number of individuals of the same species appearing simultaneously in any single frame of the corresponding sampling window. It is used to characterize the relative abundance under a fixed camera field of view and does not represent the absolute number of individuals of that species within the monitoring area. As shown in Table 1, Table 1 uses crucian carp as an example to list the descriptive statistics of MaxN under different candidate sampling durations.
[0034] Table 1: MaxN descriptive statistics of crucian carp under different candidate sampling durations As the sampling duration increases, the number of images contained in a single time window increases, and the chance of observing a larger number of individuals within a frame increases accordingly. Therefore, the MaxN empirical distribution may change with the sampling duration, and it is not advisable to select a single sampling duration solely based on experience.
[0035] S6. Based on the MaxN value sequence, compare the empirical distribution of MaxN corresponding to different candidate sampling durations, evaluate the change and stability of MaxN with the increase of sampling duration, and determine the recommended sampling duration for each species. For each target species, the MaxN corresponding to six candidate sampling durations is summarized. This embodiment employs a rank-based nonparametric multiple comparison method, setting the candidate sampling duration as the grouping factor, using a two-tailed test, and setting up Tukey-type pairwise comparisons across all groups. During the analysis, MaxN is transformed using log10(MaxN+1); since this transformation preserves the rank relationship of the data, it does not change the rank-based nonparametric comparison results. The statistical program outputs Comparison (comparison group), Estimator (effect estimate), Lower (lower limit), Upper (upper limit), Statistic (statistic), and p.Value (P-value), where Estimator represents the relative comparative effect estimate between the two candidate sampling durations, Lower and Upper represent the lower and upper limits of the corresponding simultaneous confidence intervals, Statistic represents the test statistic, and p.Value represents the p-value after multiple comparison correction based on the joint distribution of the comparison statistics. Multiple comparison letter groups are generated based on the corrected p-values; the same letter indicates no significant difference was detected at the preset significance level, while different letters indicate a significant difference was detected.
[0036] In this embodiment, the significance level is set to 0.05. When the p-value after multiple comparison correction between a candidate sampling duration and all longer candidate sampling durations is not less than 0.05, the shortest candidate sampling duration that meets the above condition is determined as the recommended sampling duration for the corresponding species; if no candidate sampling duration meets the above condition, the longest candidate sampling duration is determined as the recommended sampling duration. In the table, p.Value represents the p-value after multiple comparison correction output by the statistical program; a value of 0.000000e+00 indicates that the corresponding p-value is lower than the numerical output precision of the software, and does not mean that the theoretical probability is strictly equal to 0.
[0037] (1) Determination of recommended sampling time for crucian carp; The pairwise comparison results of crucian carp among the six candidate sampling durations are shown in Table 2.
[0038] Table 2: Pairwise comparison results of crucian carp samples with different candidate sampling durations The results of multiple comparisons of crucian carp are shown in Table 3: Table 3: Multiple comparison groups of crucian carp samples under different candidate sampling durations The estimated value for the comparison between the 720-minute and 1440-minute sampling durations for crucian carp was 0.603, with a confidence interval of 0.429–0.754, a statistical statistic of 1.670697, and a p-value of 0.5494267, which is greater than 0.05. Both sampling durations were classified into group e, indicating that no significant difference was detected between the 720-minute and 1440-minute sampling results under the statistical test rules of this embodiment. Meanwhile, the p-values between 360 minutes and 720 minutes were 0.02848005, and between 360 minutes and 1440 minutes were 0.0001050708, both less than 0.05, and 360 minutes was classified into group d. Therefore, candidate sampling durations shorter than 720 minutes did not meet the stability criteria, and no significant difference was detected between 720 minutes and the only longer candidate sampling duration of 1440 minutes. Therefore, 720 minutes was determined as the recommended sampling duration for crucian carp.
[0039] (2) Determination of recommended sampling duration for Chinese medaka The pairwise comparison results of the Chinese medaka among the six candidate sampling durations are shown in Table 4: Table 4: Pairwise comparison results of Chinese medaka between different candidate sampling durations The estimated value for the comparison between sampling durations of 720 minutes and 1440 minutes for *Phyllostachys sinensis* was 0.724, with a confidence interval of 0.558–0.845, a statistical value of 3.742004, and a p-value of 0.001979255, which is less than 0.05, indicating a significant difference between the 720-minute and 1440-minute sampling results. Furthermore, the p-values for all 15 pairwise comparisons were less than 0.05. Therefore, among the five candidate sampling durations (30 minutes, 60 minutes, 180 minutes, 360 minutes, and 720 minutes), there were no candidate values that did not show a significant difference from any of the longer candidate sampling durations. According to the pre-defined decision rules, the longest candidate sampling duration of 1440 minutes was determined as the recommended sampling duration for *Phyllostachys sinensis*.
[0040] (3) Comparison of recommended sampling durations for different species; The key comparison results and recommended sampling durations for crucian carp and Chinese medaka are shown in Table 5.
[0041] Table 5: Comparison of recommended sampling duration for crucian carp and Chinese medaka The above results indicate that the MaxN response to sampling duration is inconsistent among different fish species. For crucian carp, no significant difference was detected when the sampling duration was extended from 720 minutes to 1440 minutes, and 720 minutes is sufficient to meet the stability criteria of this embodiment. For Chinese medaka, a significant difference still exists between 720 minutes and 1440 minutes, and 1440 minutes should be used as the recommended sampling duration. Therefore, determining the recommended sampling duration for different species helps reduce the risk of mismatch caused by uniformly adopting empirical sampling durations and provides species-specific sampling basis for the design of fixed underwater video monitoring schemes.
[0042] S7. Construct a relative abundance assessment window based on the recommended sampling duration, calculate the MaxN of each species within the relative abundance assessment window, and use MaxN to characterize the relative abundance of each species under the conditions of fixed camera field of view and consistent video acquisition. Finally, output the recommended sampling duration and relative abundance assessment results for each species.
[0043] The relative abundance assessment windows were reconstructed according to the recommended sampling duration for each species. Specifically, a 720-minute window was constructed for crucian carp, and a 1440-minute window was constructed for Chinese medaka. The MaxN for the corresponding species within each assessment window was calculated, and this MaxN was used as the relative abundance index for that species to compare the relative abundance changes between different monitoring dates or periods under conditions of fixed camera equipment, fixed camera field of view, and consistent video acquisition. MaxN is not used to represent the absolute number of fish individuals within the monitoring area.
[0044] After the operation is completed, the output includes frame-by-frame species count records for each species, MaxN results for different candidate sampling durations, MaxN empirical distribution map, nonparametric multiple comparison results based on Tukey-type comparisons, recommended sampling duration, and relative abundance assessment results. It also simultaneously saves the target detection model version, species confidence threshold, candidate sampling duration, sliding step size, significance level, statistical test methods, and multiple comparison parameters, such as... Figure 4 As shown.
[0045] This embodiment can be used in wetland waters such as rivers, lakes, reservoirs, ponds and swamps. It can carry out fish video monitoring under a fixed camera field of view in a low-interference, non-contact and non-invasive manner, and provide a basis for determining the appropriate sampling duration for fish relative abundance surveys, assessing changes in fish relative abundance, formulating long-term underwater video monitoring plans and optimizing video data processing workload.
[0046] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the described technical solutions or make equivalent substitutions for some of the technical features; these modifications or substitutions do not cause the substance of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video, characterized in that, Specifically, the following steps are included: S1. By deploying fixed underwater camera equipment, continuously collect videos of fish activities and record site information, equipment information, and time information; S2. Construct a list of monitored species and a basic image database of fish; S3. Manually annotate the bounding boxes and species categories of fish targets in the basic fish image database, and divide the annotated images into independent training, validation and test sets according to a preset ratio. Train the object detection model to obtain a well-trained object detection model; S4. Based on the acquired continuous video and the target detection model, extract image frames at preset fixed time intervals, run the target detection model on each image frame, output the fish target bounding box, species, confidence score and detection count, generate a frame-by-frame species count record, and output a structured frame-by-frame species count record. S5. Based on the frame-by-frame species counting record, construct a time window according to a preset number of candidate sampling durations, calculate the maximum number of individuals MaxN of each species that appear simultaneously in any single frame image within each time window, and obtain the MaxN value sequence of each species in each time window under each candidate sampling duration. S6. Based on the MaxN value sequence, compare the empirical distribution of MaxN corresponding to different candidate sampling durations, evaluate the change and stability of MaxN with the increase of sampling duration, and determine the recommended sampling duration for each species. S7. Construct a relative abundance assessment window based on the recommended sampling duration, calculate the MaxN of each species within the relative abundance assessment window, and use MaxN to characterize the relative abundance of each species under fixed camera field of view and consistent video acquisition conditions. Finally, output the recommended sampling duration and relative abundance assessment results for each species.
2. The method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video as described in claim 1, characterized in that, In step S2, the construction of the monitoring species list and the basic fish image database specifically involves: based on historical survey data and on-site pre-survey results, determining the potential fish species distributed within the monitoring area, and identifying the dominant species, common species, and key monitoring species to form a monitoring area fish species list; for the target monitoring species in the fish species list, collecting fish images covering different lighting conditions, water turbidity, individual posture, individual size, and degree of occlusion to establish a basic fish image database for the monitoring area.
3. The method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video as described in claim 1, characterized in that, In step S3, the target detection model outputs the fish target bounding box, species category, confidence score, and number of detections for each frame of image, and uses the test set to calculate one or more of precision, recall, F1 score, mean precision, or mean mean precision to evaluate the recognition performance of the target detection model.
4. The method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video as described in claim 1, characterized in that, In step S5, the candidate sampling durations constitute the following set: For date and candidate sampling duration Based on the base time of that date Starting from the sliding step size Construct the k-th time window: Candidate sampling duration within the corresponding date The corresponding number of time windows is: In the formula, For the candidate sampling duration set, Let m be the sampling duration of the candidate. Indicates the number of candidate sampling durations. Indicates the first Candidate sampling duration, and These represent the start and end times of monitoring on date d, respectively. Indicates the candidate sampling duration The corresponding sliding step size, and 0 < ≤ , Indicates the date d during the candidate sampling duration. The k-th time window below, Indicates the total number of corresponding time windows; when = When 0 < 0, adjacent time windows do not overlap; when 0 < 0, adjacent time windows do not overlap. < At that time, adjacent time windows partially overlap.
5. The method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video as described in claim 4, characterized in that, In step S6, the expression for calculating MaxN for each species within the relative abundance assessment window is as follows: In the formula, Indicates species s within a time window The maximum number of individuals that can appear simultaneously in any single frame of an image; This indicates that the timestamp of the i-th frame is within the time window. This represents the timestamp of the i-th frame within date d. This represents the number of species s detected in the i-th frame image; When species s is not detected in the i-th frame image, Record it as 0; allocate the frame-by-frame species count records to the corresponding time windows according to the image timestamp, and calculate the MaxN of species s in the date d, the sampling duration of the j-th candidate and the k-th time window according to the following formula.
6. The method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video as described in claim 5, characterized in that, MaxN is used to characterize the relative abundance of species s within the corresponding sampling window under a fixed camera field of view, and does not represent the absolute number of individuals of species s in the monitoring area.
7. The method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video as described in claim 1, characterized in that, In step S6, comparing the empirical distributions of MaxN corresponding to different candidate sampling durations specifically involves: summarizing the MaxN obtained for each species under different candidate sampling durations within a preset monitoring period to form MaxN sample sets corresponding to each candidate sampling duration; for the same species, using a rank-based nonparametric multiple comparison method and setting up Tukey-type pairwise comparisons to compare the MaxN distributions corresponding to different candidate sampling durations, obtaining the relative contrast effect estimate and the P-value corrected by multiple comparisons; when the corrected P-value is less than the preset significance level, it is determined that there is a significant difference in the MaxN distributions corresponding to the two candidate sampling durations. When the corrected P-value is not less than the preset significance level, it is determined that no significant distributional difference was detected under the statistical test rule.
8. The method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video as described in claim 1, characterized in that, In step S6, determining the recommended sampling duration for each species specifically involves: placing species s in the candidate sampling duration... and The p-value of a pairwise Tukey-type comparison, after multiple comparison correction, is denoted as Let the preset significance level be denoted as α, and determine the recommended sampling duration for species s according to the following formula: When there are no candidate sampling durations that satisfy the above conditions, let: In the formula, Indicates the recommended sampling duration for species s. Indicates the number of candidate sampling durations. Indicates the first Candidate sampling duration, This represents the p-value after multiple comparison correction, and α represents the pre-set significance level. and An index for the candidate sampling duration. Indicates the first Candidate sampling duration, The sampling duration for the m-th candidate; When the corrected P-value between a candidate sampling duration and all longer candidate sampling durations is not less than α, it is determined that no significant distribution difference was detected under the preset statistical test rules, and the MaxN empirical distribution of the species tends to stabilize with the increase of sampling duration. The shortest candidate sampling duration that meets the above conditions is determined as the recommended sampling duration for this species.
9. The method for assessing the relative abundance of fish in wild wetlands based on fixed underwater video as described in claim 1, characterized in that, In S7, the relative abundance index is determined according to the following formula: In the formula, This represents the relative abundance index of species s within the k-th relative abundance assessment window on date d. For the monitoring date, Here, s represents the index of the relative abundance assessment window, and s represents the target species. Indicates the recommended sampling duration for species s; The relative abundance index is used to compare the relative abundance changes between different monitoring dates or monitoring periods under conditions of fixed camera equipment, fixed camera field of view, and consistent video acquisition.
10. A system for assessing the relative abundance of fish in wild wetlands based on fixed underwater video, used to perform the method described in any one of claims 1 to 8, characterized in that, Includes fixed underwater video acquisition devices and data processing devices; The data processing device includes a frame-by-frame species identification and counting module, a candidate sampling duration setting module, a time window construction module, a MaxN calculation module, a MaxN empirical distribution construction and comparison module, a recommended sampling duration determination module, and a relative abundance evaluation module. The time window construction module is used to group the frame-by-frame species counting records by time according to the candidate sampling duration and the sliding step size. The MaxN calculation module is used to calculate the maximum number of individuals of each species that appear simultaneously in any single frame image within each time window. The relative abundance assessment module is used to construct a relative abundance assessment window according to the recommended sampling duration for each species, and the MaxN within the corresponding window represents the relative abundance of each species.