Intelligent benthic organism identification system

Abstract

This invention discloses an intelligent identification system for benthic organisms. In the field of artificial intelligence technology, this system quantitatively groups benthic organisms according to their impact on water quality assessment based on the Hilsenhoff pollution tolerance system. Differentiated feature extraction priorities, counting weights, and confidence thresholds are set for different groups. After multi-module collaborative processing, the system combines the pollution tolerance values ​​to complete a standardized evaluation of water quality levels. The advantages of this invention are: it quantitatively groups benthic organisms according to their impact on water quality assessment based on the Hilsenhoff pollution tolerance system, abandoning the traditional method of indiscriminately optimizing the accuracy of all species identification. It implements targeted processing for different groups, including differentiated feature extraction priorities, differentiated counting weights, and differentiated confidence threshold verification. It focuses on core water quality indicator species to achieve key optimization of identification accuracy, directly preventing errors in species identification and counting from being transmitted to water quality evaluation results at the technological source.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a benthic organism intelligent identification system. Background Technology

[0002] Benthic organisms, as an important component of aquatic ecosystems, are closely related to water quality through their community structure, species distribution, and abundance variations. They serve as core indicator organisms of water quality, and benthic monitoring and evaluation have become crucial technical means in the field of aquatic ecological environment monitoring, playing an irreplaceable role in watershed ecological protection, water quality assessment, and pollution control effectiveness detection. With the deepening of ecological civilization construction in my country, the requirements for timeliness, accuracy, and on-site application in water environment monitoring are constantly increasing. Therefore, rapid identification and quantitative analysis of benthic organisms have become an important research direction in the field of aquatic ecological environment monitoring.

[0003] Existing intelligent benthic organism identification systems mainly achieve benthic organism identification, counting, and water quality assessment by indiscriminately extracting species visual features, uniformly configuring counting weights, and setting verification thresholds. This approach has certain drawbacks. First, it does not categorize species according to the Hilsenhoff pollution tolerance system, and indiscriminately optimizes the identification accuracy of all species. The indiscriminate transmission of identification and counting errors leads to distorted water quality assessment results. Second, indiscriminately improving the identification accuracy of all species causes a surge in computing power requirements. This presents a technical bottleneck: if the identification accuracy meets the standards, it cannot be deployed on mobile devices; if it can be deployed on-site, the water quality assessment will not be compliant. Therefore, we propose an intelligent benthic organism identification system. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent identification system for benthic organisms.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a benthic organism intelligent identification system, the intelligent identification system comprising the following modules:

[0006] Image acquisition guidance module: This module provides standardized operational guidelines for the field acquisition of benthic organism images, performs basic quality screening on the acquired images, and outputs standardized benthic organism images that meet the requirements for subsequent identification.

[0007] Species grouping module: Based on the Hilsenhoff pollution tolerance value system, it quantitatively groups benthic organisms according to their impact weight on water quality assessment results, and outputs benthic organism grouping results containing group information and corresponding impact weights;

[0008] Feature extraction and recognition module: It is used to receive standardized benthic organism images and group classification results, set differentiated feature extraction priorities for benthic organisms in different groups, allocate more computing resources to high-impact weight groups and prioritize the extraction of their key visual features, complete species identification and individual counting, and output benthic organism species identification results and corresponding original individual count results;

[0009] The counting weight allocation module receives species identification results, raw individual count results, and group division results. Based on the influence weight corresponding to the group, it configures differentiated counting weights for the raw individual count results of benthic organisms in different groups and outputs the weighted benthic organism individual count results.

[0010] Error verification module: It is used to receive the weighted individual count results and group division results, set differentiated species identification confidence verification thresholds for different groups, complete the error verification of the count results, and output the benthic organism individual count results that pass the verification.

[0011] Water quality assessment module: It is used to receive the verified individual count results and group classification results, combine them with the benthic organism pollution tolerance value to complete the standardized evaluation of water quality level, and output the water quality assessment results corresponding to the benthic organism identification.

[0012] As a further aspect of the present invention: in the species grouping module, the specific method for grouping benthic organisms based on the Hilsenhoff pollution tolerance system is implemented through a grouping formula, which is as follows:

[0013] ;

[0014] in, For the first Group identifiers for benthic organisms. For the first Hilsenhoff tolerance values ​​for benthic organisms This is a core sensitive category, which consists of key indicator species for water cleanliness and has a high weighting in influencing water quality assessment results. This is the core pollution-tolerant group, which consists of key indicator species for water pollution and has the second highest weighting in influencing water quality assessment results. As a non-core impact group, the pollution tolerance value of benthic organisms in this group is in the middle range, and has no significant impact on the water quality evaluation results. The impact weight is the lowest. This module will associate and store the above group identifiers, pollution tolerance values ​​and corresponding impact weights to form the benthic organism group classification results and transmit them to the subsequent modules.

[0015] As a further aspect of the present invention: in the feature extraction and recognition module, the differentiated feature extraction priorities set for different groups of benthic organisms are realized through a feature extraction weight formula, which is as follows:

[0016] ;

[0017] In the formula, For the first Feature extraction weights for benthic organisms. For group number, 1, 2, 3 correspond to respectively , , , The normalization coefficient for feature extraction weights, with a value of [value to be filled in]. Used to guarantee This module allocates corresponding GPU computing resources based on feature extraction weights. The higher the weight, the larger the proportion of computing resources allocated. It prioritizes the extraction of key visual features, species matching, and individual counting of benthic organisms in high-weight groups, and then performs identification and counting of low-weight groups to ensure the identification accuracy of core indicator species.

[0018] As a further aspect of the present invention: in the counting weight allocation module, the differentiated counting weights configured for different groups of benthic organisms are achieved through a counting weight assignment formula, which is as follows:

[0019] ;

[0020] In the formula, For the first Individual count weights for benthic organisms in the group. 1, 2, 3 correspond to respectively , , , The counting weight calibration coefficient, with a value between 0 and 1, is a fixed calibration value adapted to the water quality monitoring industry standard. The formula used by the counting weight allocation module to calculate the weighted individual count result is as follows: ,in For the first Weighted count of benthic organisms For the first This module provides the original individual count results for benthic organisms and compares the species identification results of all benthic organisms with their corresponding species. The results are correlated to form a weighted count of benthic organisms and then transmitted to the error verification module.

[0021] As a further aspect of the present invention: in the error verification module, the differentiated species identification confidence verification threshold set for different groups is implemented through a threshold setting formula, which is:

[0022] ;

[0023] In the formula, For the first Species identification confidence threshold for benthic organisms. 1, 2, 3 correspond to respectively , , , This is the threshold baseline coefficient, ranging from 0 to 1. It serves as a fixed baseline value adapted to field image recognition scenarios. When the species identification confidence value of a certain benthic organism is lower than that of its corresponding group... At that time, the error verification module triggers secondary feature extraction and species identification verification of the benthic organism, and the confidence level reaches [a certain threshold] after verification. The weighted individual count results are retained, and the confidence level is still lower than [previous level] after review. The count result is then discarded, and only the weighted individual count results that pass the review or meet the initial confidence level are transmitted to the water quality assessment module as the valid benthic individual count results.

[0024] As a further aspect of the present invention: the standardized operation guidelines of the image acquisition guidance module include background selection guidelines, shooting distance guidelines, lighting condition guidelines, and shooting mode guidelines. The background selection guidelines require placing benthic organisms on a white or light-colored, textureless background. The shooting distance guidelines require maintaining a vertical distance of 10-15cm between the shooting device and the benthic organisms. The lighting condition guidelines require sufficient ambient light and no obvious shadows. The shooting mode guidelines require using macro shooting mode. This module has built-in image quality screening rules to quantitatively detect the background interference, sharpness, and illumination uniformity of the acquired images, and removes unqualified images that do not meet the above guidelines. Only qualified images are transmitted to the species grouping module as standardized benthic organism images. At the same time, the positioning information of the shooting device is collected and stored, and the positioning information is synchronously transmitted to the GIS data linkage module.

[0025] As a further aspect of the present invention: the water quality assessment module incorporates a standardized benthic organism pollution tolerance database and a water quality grade assessment system. The pollution tolerance database is associated with the group classification results and stores the names, Latin names, and corresponding Hilsenhoff pollution tolerance values ​​of each benthic species. The water quality grade assessment system is divided into four levels: clean, lightly polluted, moderately polluted, and heavily polluted. The water quality assessment module calculates the water quality assessment index based on the count results of qualified individuals and the pollution tolerance database. Then, it matches the corresponding water quality grade based on the calculation results to form a water quality assessment result that includes species identification summary, count result summary, water quality assessment index, and water quality grade. At the same time, it outputs corresponding water ecological protection recommendations.

[0026] As a further aspect of the present invention: the intelligent identification system further includes a GIS data linkage module and an identification record storage module. The GIS data linkage module is used to receive the water quality evaluation results and the collection location information transmitted by the image acquisition guidance module, bind and fuse the water quality evaluation results with geospatial information, and output the water quality evaluation results with geographic information.

[0027] The GIS data linkage module receives location information including latitude and longitude, altitude, watershed affiliation, and administrative division information. This module has a built-in geographic information system engine that structurally binds the water quality assessment results with the aforementioned location information, generating visualized water quality assessment results with geographic coordinates. It also supports exporting the visualized data as a vector map. This module synchronously transmits the water quality assessment results with geographic information to the identification record storage module, realizing the spatial display and management of water quality assessment results.

[0028] As a further aspect of the present invention: the identification record storage module is used to receive standardized benthic organism images, species identification results, verified individual count results, water quality evaluation results, and water quality evaluation results with geographic information, and to complete the classification, storage, retrieval, and statistical management of the entire process data. The feature extraction and identification module, the counting weight allocation module, and the error verification module form a collaborative linkage mechanism based on the grouping results of the species grouping module. Through differentiated feature extraction, differentiated counting weight allocation, and differentiated error verification, the error in the species identification and counting process is prevented from being transmitted to the water quality evaluation results at the source.

[0029] The identification record storage module uses a MySQL database for data storage. The database is divided into image storage sub-database, species identification sub-database, counting result sub-database, water quality evaluation sub-database, and geographic information sub-database according to data type. Each sub-database establishes an associated retrieval index, supporting single or multiple conditional searches based on collection time, collection location, benthic species, and water quality level. This module also has data statistical analysis functions, which can automatically statistically analyze the distribution, quantity changes, and water quality level changes of benthic species within a specified time range and area, generate statistical analysis reports, and support exporting the entire process data to standardized formats such as Excel and PDF.

[0030] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0031] 1. This invention constructs a technical system based on species grouping, with deep collaboration between feature extraction and identification modules, counting weight allocation modules, and error verification modules. Based on the Hilsenhoff pollution tolerance system, benthic organisms are quantitatively grouped according to their impact weight on water quality assessment. It abandons the traditional method of indiscriminately optimizing the accuracy of all species identification in existing technologies. Instead, it implements targeted processing for different groups, such as differentiated feature extraction priority settings, differentiated counting weight configurations, and differentiated confidence threshold verification. It focuses on core water quality indicator species to achieve key optimization of identification accuracy. It directly blocks the transmission of errors in species identification and counting to water quality assessment results from the source of technology. It solves the inherent defect of existing technologies that cause distortion of water quality assessment results due to the indiscriminate transmission of identification errors. Finally, it achieves a high degree of consistency between the water quality assessment results output from the field and the laboratory standard test results, ensuring the compliance and accuracy of water quality assessment results in aquatic ecological monitoring.

[0032] 2. This invention designs the functional logic and data transmission relationships of each module of the system with water quality assessment compliance as the core guiding principle. Based on the species grouping results, each module allocates more computing resources only to species that have a core impact on water quality assessment, without indiscriminately improving the identification accuracy of all species. This avoids the problem of a surge in system computing power demand caused by high-precision optimization of all species in existing technologies, and significantly reduces the overall computing power consumption of the system. It fundamentally solves the technical bottleneck of existing technologies where the identification accuracy meets the standard but cannot be deployed on mobile devices, or where the water quality assessment results are non-compliant when deployed on-site. It realizes lightweight deployment of the system in the field on low-computing-power mobile devices such as mobile phones, and achieves a closed-loop technology for real-time on-site identification of benthic organisms and compliant water quality assessment, meeting the actual needs of rapid on-site monitoring of aquatic ecosystems. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the system flow in an embodiment of the present invention. Detailed Implementation

[0034] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0035] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0036] Please see the appendix Figure 1 The present invention discloses a benthic organism intelligent identification system, which includes the following modules:

[0037] Image acquisition guidance module: This module provides standardized operational guidelines for the field acquisition of benthic organism images, performs basic quality screening on the acquired images, and outputs standardized benthic organism images that meet the requirements for subsequent identification.

[0038] Species grouping module: Based on the Hilsenhoff pollution tolerance value system, it quantitatively groups benthic organisms according to their impact weight on water quality assessment results, and outputs benthic organism grouping results containing group information and corresponding impact weights;

[0039] Feature extraction and recognition module: It is used to receive standardized benthic organism images and group classification results, set differentiated feature extraction priorities for benthic organisms in different groups, allocate more computing resources to high-impact weight groups and prioritize the extraction of their key visual features, complete species identification and individual counting, and output benthic organism species identification results and corresponding original individual count results;

[0040] The counting weight allocation module receives species identification results, raw individual count results, and group division results. Based on the influence weight corresponding to the group, it configures differentiated counting weights for the raw individual count results of benthic organisms in different groups and outputs the weighted benthic organism individual count results.

[0041] Error verification module: It is used to receive the weighted individual count results and group division results, set differentiated species identification confidence verification thresholds for different groups, complete the error verification of the count results, and output the benthic organism individual count results that pass the verification.

[0042] Water quality assessment module: It is used to receive the verified individual count results and group classification results, combine them with the benthic organism pollution tolerance value to complete the standardized evaluation of water quality level, and output the water quality assessment results corresponding to the benthic organism identification.

[0043] Example 1

[0044] This embodiment was applied to the field monitoring and water quality assessment of benthic organisms in a clean tributary of a lake. The water cleanliness of this basin is relatively high, and the core sensitive species among the benthic organisms ( The proportion of these areas is relatively high, the lighting conditions are good, the aquatic environment is simple, and there is no obvious background interference, making them suitable for rapid on-site monitoring in the field.

[0045] In this embodiment, all modules of the system are deployed on an Android mobile device, equipped with a high-definition macro camera and a portable GIS positioning module. On-site, the system completes benthic organism image acquisition, identification, counting, and water quality assessment. The specific operation steps are as follows:

[0046] Standardized processing of the image acquisition guidance module:

[0047] The data collection was performed according to the standardized operating instructions built into the module: Benthic organisms were placed on a white, textureless glass slide, and the imaging device was kept 12cm vertically from the sample. Natural light (without shadows) was used, and macro shooting mode was activated. The image acquisition guidance module screened the acquired images for quality, quantitatively detecting background interference (≤5%), sharpness (pixel resolution ≥300dpi), and illumination uniformity (deviation ≤10%). Three unqualified images were removed, and 120 standardized benthic organism images were finally output. At the same time, the GIS location information of the watershed (latitude and longitude: 120.35°E, 31.12°N, altitude 3.2m, watershed belongs to Taihu Lake Basin, administrative division is Suzhou City, Jiangsu Province) was collected and stored, and synchronously transmitted to the GIS data linkage module.

[0048] Quantitative grouping of species:

[0049] The species grouping module is based on the Hilsenhoff pollution tolerance system. It uses a grouping formula to classify the collected benthic organisms into groups. In this example, a total of 15 benthic organisms were collected, among which... (Core sensitive class, ≤3) 5 species (such as mayfly nymphs) =1.5, Caddisfly nymphs =2.0 etc.), (Core stain-resistant category, ≥7) 3 types (such as midge larvae) =7.5, Waterworm =8.0 etc.), (Non-core impact category, 3 <) <7) 7 types (such as snails) =4.5, Clams =5.0 etc.), the species grouping module stores the group identifier, pollution tolerance value, and influence weight of each species in association, among which The one with the highest influence weight Second highest At the lowest level, the group division results are generated and transmitted to subsequent modules.

[0050] Differential feature extraction and recognition counting in the feature extraction and recognition module:

[0051] The feature extraction and recognition module receives standardized images and grouping results, and calculates the feature extraction weights for different groups using the feature extraction weight formula: Substitute into the calculation: =1 / (1+1 / 2+1 / 3)=6 / 11≈0.545, therefore =6 / 11≈0.545、 =3 / 11≈0.273、 =2 / 11≈0.182, the feature extraction and recognition module allocates GPU computing resources according to the weights: 54.5% allocated 27.3% allocated 18.2% allocated, priority extraction , The key visual features of the core species (such as body shape, appendage morphology, and body surface texture) are used to complete species matching and individual counting, and finally output the species identification results and the original individual count results. The total original count for the group was 185. The total original count for the group was 62. The total original count for the group was 213.

[0052] Weighted count calculation in the counting weight allocation module:

[0053] The counting weight allocation module receives species identification, raw counting, and group division results, calculates the counting weight for different groups using the counting weight assignment formula, and substitutes the results into the calculation: =0.2×3=0.6、 =0.2×2=0.4、 =0.2×1=0.2, then calculate the weighted count using the weighted counting formula, and finally output the weighted individual count result: The total weighted count for the group is 185 × 0.6 = 111. The total weighted count for the group is 62 × 0.4 = 24.8. The total weighted count for the group is 213 × 0.2 = 42.6, and the total weighted count is 178.4.

[0054] Differential confidence level verification of the error verification module:

[0055] The error verification module receives the weighted counting results and the group classification results, and sets the recognition confidence verification threshold for different groups using the threshold setting formula: , The threshold baseline coefficient is set to a fixed value of 0.9 for field scene recognition scenarios. Substituting this value into the calculation: T( ) = 0.9 × 1 = 0.9, T( ) = 0.9 × 0.9 = 0.81, T( =0.9 × 0.8 = 0.72. The error verification module verifies the confidence level of each species identification: It was found that... The confidence level for identifying one mayfly nymph species in the group was 0.85 (below the threshold of 0.9), triggering secondary feature extraction and species identification verification. After verification, the confidence level increased to 0.93, and its weighted count result was retained. The confidence levels for the remaining species all met the standard, and the final output was a qualified weighted count result. Group 111 Group 24.8 Group 42.6, total count 178.4).

[0056] The water quality assessment module is linked with GIS data, and subsequent processing is performed by the identification, recording, and storage module.

[0057] The water quality assessment module calls upon the built-in pollution tolerance database and water quality grade assessment system, and calculates the water quality assessment index by combining the weighted count results of the qualified data. It then matches the water quality grade of the watershed to clean and outputs water ecological protection recommendations (maintain watershed biodiversity and reduce agricultural non-point source pollution in the surrounding area).

[0058] The GIS data linkage module binds and integrates water quality assessment results with collected GIS positioning information to generate a visualized vector map with geographic coordinates.

[0059] The identification record storage module classifies and stores standardized images, species identification results, verified count results, water quality assessment results, and geographic information into various sub-databases of the MySQL database, establishes an associated retrieval index, and supports multi-condition retrieval.

[0060] Example 2

[0061] This embodiment was applied to the field monitoring and water quality assessment of benthic organisms in a mining watershed in Southwest China. The watershed was slightly affected by industrial wastewater, resulting in a lightly polluted state. The core pollution-tolerant benthic organisms (…) The proportion of [unclear] is relatively high, and the field scenes are complex (insufficient lighting, large background interference, rugged watershed terrain), requiring high image acquisition quality and system anti-interference capabilities. In this embodiment, the system is also deployed on the aforementioned Android mobile terminal, equipped with a high-definition macro camera + portable supplementary lighting device and a high-precision GIS positioning module. The specific operating steps are basically the same as in Embodiment 1. The adaptation adjustments for complex scenes and the core results are as follows:

[0062] Adaptation processing of the image acquisition guidance module: Due to insufficient on-site lighting, a portable supplementary lighting device (light intensity ≥ 500 lux) was turned on, and the shooting distance was adjusted to 10cm to reduce background interference. The module acquired a total of 150 images, discarded 12 unqualified images, and output 138 standardized images. At the same time, GIS positioning information was acquired (latitude and longitude: 103.52°E, 25.48°N, altitude 856.3m, watershed belongs to the upper reaches of the Yangtze River tributary, and the administrative division is Qujing City, Yunnan Province).

[0063] Species group classification results: A total of 18 benthic organisms were collected, among which... ( ≤3) 4 types ( ≥7) 7 types (such as trembling earthworms) =7.8, *Bugnae* =8.5 etc.), (3< <7) 7 types, with the proportion of group G2 being significantly higher than that of Example 1.

[0064] Feature extraction and counting weight optimization: Normalization coefficients for feature extraction We still take 6 / 11, because This group represents the core indicator species of the watershed, and the module allocates GPU resources accordingly. The group performed a slight tilt ( 53% 29% :18%); Count weight calibration coefficient Still set at 0.2, threshold baseline coefficient The value remains 0.9, and the rules for calculating the differential threshold and weights are the same as in Example 1.

[0065] Core identification and evaluation results: The original counting results are as follows Group 102 Group 205 Group 198; weighted count result is Group 61.2 Group 82 Group 39.6; Triggered during error check The two species in the group underwent a second review, and both met the standards after the review. The water quality assessment module calculated that the water quality level of the basin was slightly polluted, and output protection recommendations (strictly control the discharge of industrial and mining wastewater, carry out watershed ecological restoration, and monitor changes in benthic community structure).

[0066] Subsequent processing: The GIS data linkage module performs watershed subdivision visualization based on the characteristics of the mountainous watershed, and the identification, recording and storage module completes the data storage of the entire process and generates a statistical report on the distribution of benthic species in the watershed.

[0067] Comparative Example

[0068] This comparative example uses a traditional, undifferentiated benthic organism intelligent identification system. This system does not classify species based on the Hilsenhoff pollution tolerance system, and performs undifferentiated feature extraction, undifferentiated counting weighting, and the same confidence level verification threshold for all benthic species. The specific processing rules are as follows:

[0069] Feature extraction: GPU computing resources are allocated equally to all species (each species has an equal share), with no extraction priority;

[0070] Count weight: The count weight for all species is set to 1, and the original count is used directly as the final count result;

[0071] Error verification: The confidence threshold for identification of all species is set to 0.8, and there are no differentiated triggering rules for secondary verification;

[0072] The remaining modules (image acquisition, water quality assessment, GIS linkage, etc.) are the same as in Example 1.

[0073] Comparative Example 2

[0074] This comparative example is a semi-differentiated improved traditional benthic organism intelligent identification system. This system only performs differentiation in feature extraction, without differentiating the configuration of counting weights or setting differentiating thresholds for error verification. The specific processing rules are as follows:

[0075] Feature extraction: Consistent with Example 1, groups were divided and GPU computing resources were allocated based on Hilsenhoff stain tolerance values;

[0076] Count weights: All species are assigned a count weight of 1, and no weighted counts are calculated.

[0077] Error verification: The confidence threshold for identification of all species was set to 0.8, indicating no difference.

[0078] The remaining modules are the same as in Example 1.

[0079] Experimental verification

[0080] Experimental equipment and experimental samples

[0081] Experimental equipment: Android mobile device (Snapdragon 888, 8GB RAM), 48MP high-definition macro camera, portable GIS positioning module, portable lighting equipment, MySQL database server;

[0082] Experimental Samples: Example 1 and Comparative Examples 1-2 used benthic organism samples (500 individuals, 15 species, including...) collected in the field from a clean tributary of Taihu Lake. 5 types 3 types 7 species); Example 2 uses benthic organism samples (505 individuals, 18 species, including 7 species) collected in the field in a mining area in Southwest China. 4 types 7 types 7 types);

[0083] Laboratory standard data: All field-collected samples were brought back to the laboratory, and species identification and counting were completed using morphological identification methods. The water quality evaluation index was calculated using the Shannon-Wiener index method combined with the Hilsenhoff pollution tolerance value as the standard true value.

[0084] Experimental verification indicators

[0085] The core evaluation indicators selected for field monitoring cover three dimensions: identification accuracy, water quality assessment accuracy, and system computing power consumption. The specific indicators are as follows:

[0086] Core species ( + ) Identification confidence level: Reflects the accuracy of the system in identifying the core species of water quality indicators. It ranges from 0 to 1, with higher values ​​indicating higher accuracy.

[0087] Overall species identification accuracy: Overall species correct identification number / total number of identifications × 100%, reflecting the overall identification capability of the system;

[0088] Water quality assessment result consistency: The degree of matching between the on-site assessment results and the laboratory standard results, divided into complete consistency, basic consistency, and non-consistency. Complete consistency is recorded as 100%, basic consistency as 80%, and non-consistency as 0%.

[0089] Average GPU utilization: Reflects the system's computing power consumption; the lower the value, the more suitable it is for mobile deployment.

[0090] Mobile terminal average response time: The total time from image acquisition to output of water quality assessment results, reflecting the system's ability to respond quickly on-site.

[0091] Experimental Results and Analysis

[0092] Table 1. Comparison of experimental indicators for various schemes in a clean tributary watershed of a certain lake:

[0093] Evaluation indicators Example 1 (System of the Invention) Comparative Example 1 (Undifferentiated Traditional System) Comparative Example 2 (Semi-differentiated improvement system) Core species ( + ) Identification Confidence 0.92 0.75 0.88 Overall species identification accuracy 0.968 0.823 0.925 Water quality assessment results consistency 1 0 0.8 Average GPU utilization 0.452 0.897 0.586 Mobile average response time 18.5s 65.3s 32.7s

[0094] Table 1

[0095] Table 2. Performance of experimental indicators in Example 2 of a mining and industrial basin in Southwest China:

[0096] Evaluation indicators Example 2 Experimental Results Laboratory standard verification consistency Core species ( + ) Identification Confidence 90.00% / / Overall species identification accuracy 94.30% / / Water quality assessment results consistency 100.00% 100.00% completely consistent Average GPU utilization 48.70% / / Mobile average response time 22.8s / /

[0097] Table 2

[0098] Experimental Results Analysis

[0099] Regarding recognition accuracy: The confidence level of core species recognition and the overall recognition accuracy of Example 1 are significantly higher than those of Comparative Example 1 and Comparative Example 2. This is because the system of the present invention allocates more computing resources to core indicator species, prioritizes the extraction of key features, and triggers secondary verification through differential error verification, which effectively improves the recognition accuracy of core species. In contrast, Comparative Example 1 has no differential processing, and the recognition resources for core species are diluted. Comparative Example 2 only performs feature extraction differentiation and does not further optimize through counting and verification, so the recognition accuracy is still different.

[0100] Regarding the accuracy of water quality assessment: Example 1 was 100% consistent with the laboratory standard results, Comparative Example 2 was basically consistent, and Comparative Example 1 was inconsistent. The reason is that the system of the present invention, through the collaborative process of group division, differential extraction, weighted counting, and differential verification, blocked the transmission of identification errors to water quality assessment results from the source. Moreover, the weighted counting highlighted the influence of core indicator species on water quality assessment. Comparative Example 1 did not perform differential processing, and the identification error of non-core species directly affected the assessment results. Comparative Example 2 did not perform weighted counting, and the counting weight of core species was not reflected, resulting in a deviation in the assessment results.

[0101] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Any variations and modifications can be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore, any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention, without departing from the scope of the invention, fall within the protection scope defined by the claims of the present invention.

Claims

1. A benthic organism intelligent identification system, characterized in that, The intelligent recognition system includes the following modules: Image acquisition guidance module: This module provides standardized operational guidelines for the field acquisition of benthic organism images, performs basic quality screening on the acquired images, and outputs standardized benthic organism images that meet the requirements for subsequent identification. Species grouping module: Based on the Hilsenhoff pollution tolerance value system, it quantitatively groups benthic organisms according to their impact weight on water quality assessment results, and outputs benthic organism grouping results containing group information and corresponding impact weights; Feature extraction and recognition module: Different feature extraction priorities are set for different groups of benthic organisms, more computing resources are allocated to high-influence weight groups and their key visual features are extracted first, species identification and individual counting are completed, and benthic organism species identification results and corresponding original individual count results are output; The counting weight allocation module receives species identification results, raw individual count results, and group division results. Based on the influence weight corresponding to the group, it configures differentiated counting weights for the raw individual count results of benthic organisms in different groups and outputs the weighted benthic organism individual count results. Error verification module: It is used to receive the weighted individual count results and group division results, set differentiated species identification confidence verification thresholds for different groups, complete the error verification of the count results, and output the benthic organism individual count results that pass the verification. Water quality assessment module: It is used to receive the verified individual count results and group classification results, combine them with the benthic organism pollution tolerance value to complete the standardized evaluation of water quality level, and output the water quality assessment results corresponding to the benthic organism identification.

2. The intelligent identification system for benthic organisms according to claim 1, characterized in that: In the species grouping module, the specific method for grouping benthic organisms based on the Hilsenhoff pollution tolerance system is implemented through a grouping formula, which is as follows: ； in, For the first Group identifiers for benthic organisms. For the first Hilsenhoff tolerance values ​​for benthic organisms This is a core sensitive group, which includes key indicator species for water cleanliness. This is the core pollution-tolerant group, which consists of key indicator species for water pollution. This group is classified as a non-core impact category. The pollution tolerance values ​​of benthic organisms in this group are in the middle range, and they have no significant impact on the water quality assessment results.

3. The intelligent identification system for benthic organisms according to claim 2, characterized in that: In the feature extraction and recognition module, the differentiated feature extraction priorities set for different groups of benthic organisms are implemented through a feature extraction weight formula, which is as follows: ； In the formula, For the first Feature extraction weights for benthic organisms. For group number, 1, 2, 3 correspond to respectively , , , The normalization coefficients for feature extraction weights.

4. The intelligent identification system for benthic organisms according to claim 3, characterized in that: In the counting weight allocation module, the differentiated counting weights assigned to different groups of benthic organisms are implemented through a counting weight assignment formula, which is as follows: ； In the formula, For the first Individual count weights for benthic organisms in the group. 1, 2, 3 correspond to respectively , , , The counting weight calibration coefficient, with a value between 0 and 1, is a fixed calibration value adapted to the water quality monitoring industry standard. The formula used by the counting weight allocation module to calculate the weighted individual count result is as follows: ,in For the first Weighted count of benthic organisms For the first The original individual count results of benthic organisms.

5. The intelligent identification system for benthic organisms according to claim 4, characterized in that: In the error verification module, the differentiated species identification confidence verification thresholds set for different groups are implemented through a threshold setting formula, which is: ； In the formula, For the first Species identification confidence threshold for benthic organisms. 1, 2, 3 correspond to respectively , , , This is the threshold baseline coefficient, ranging from 0 to 1. It serves as a fixed baseline value adapted to field image recognition scenarios. When the species identification confidence value of a certain benthic organism is lower than that of its corresponding group... At that time, the error verification module triggers secondary feature extraction and species identification verification of the benthic organism, and the confidence level reaches [a certain threshold] after verification. The weighted individual count results are retained, and the confidence level is still lower than [previous level] after review. Then discard the count result.

6. The intelligent identification system for benthic organisms according to claim 5, characterized in that: The standardized operation guidelines for the image acquisition guidance module include background selection guidelines, shooting distance guidelines, lighting condition guidelines, and shooting mode guidelines. The background selection guidelines require placing the benthic organisms on a white or light-colored, textureless background. The shooting distance guidelines require maintaining a vertical distance of 10-15cm between the shooting device and the benthic organisms. The lighting condition guidelines require sufficient light in the shooting environment without obvious shadows. The shooting mode guidelines require using macro shooting mode.

7. The intelligent identification system for benthic organisms according to claim 6, characterized in that: The water quality assessment module incorporates a standardized benthic organism pollution tolerance database and a water quality grading system. The pollution tolerance database is linked to the group classification results and stores the names, Latin names, and corresponding Hilsenhoff pollution tolerance values ​​of each benthic species. The water quality grading system is divided into four levels: clean, lightly polluted, moderately polluted, and heavily polluted. Based on the count results of qualified individuals, the water quality assessment module calculates the water quality assessment index in conjunction with the pollution tolerance database. Then, based on the calculation results, it matches the corresponding water quality grade to form a water quality assessment result that includes species identification summary, count result summary, water quality assessment index, and water quality grade. At the same time, it outputs corresponding aquatic ecosystem protection recommendations.

8. The intelligent identification system for benthic organisms according to claim 7, characterized in that: The intelligent recognition system also includes a GIS data linkage module and a recognition record storage module. The GIS data linkage module is used to receive the water quality evaluation results and the collection location information transmitted by the image acquisition guidance module, bind and integrate the water quality evaluation results with geospatial information, and output water quality evaluation results with geographic information.

9. The intelligent identification system for benthic organisms according to claim 8, characterized in that: The identification record storage module is used to receive standardized benthic organism images, species identification results, verified individual count results, water quality assessment results, and water quality assessment results with geographic information, and to complete the classification, storage, retrieval, and statistical management of data throughout the entire process.

Images