Intelligent benthic organism identification system
By deeply integrating convolutional neural network technology and a local database of millions of entries, the problems of low efficiency and unstable accuracy in benthic organism identification have been solved, enabling rapid and accurate identification of benthic organisms and improving the identification capabilities of aquatic ecological monitoring.
Patent Information
- Application Number
- CN202511581546.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-31
- Publication Date
- 2026-01-23
AI Technical Summary
Current technologies for identifying benthic organisms rely on human experience, resulting in low efficiency and long processing times. Traditional tools have poor regional adaptability and unstable accuracy, making it difficult to achieve fast and accurate identification.
Employing deep fusion convolutional neural network technology combined with a local database of millions of samples, this method achieves rapid and accurate identification of benthic organisms through sample collection, feature parsing, model matching, and incremental learning, demonstrating strong adaptability and generalization capabilities.
It enables rapid and accurate identification of benthic organisms, improves identification efficiency and accuracy, and solves the problems of low efficiency, unstable accuracy and poor regional adaptability in traditional methods, providing reliable technical support for aquatic ecological monitoring.
Smart Images
Figure CN121392902A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of ecological environment monitoring, in particular to an intelligent benthos identification system. BACKGROUND
[0002] Benthos refers to organisms inhabiting the bottom or bottom surface of the sea or inland waters, and is an important ecological type in aquatic organisms.
[0003] According to the way of life, it is divided into fixed life, buried life, water bottom crawling life, drilling life, bottom swimming life and other types. Such as snails, starfish, sea urchins, snake tails, etc. are the types of crawling on the seabed, and the body is often radially symmetrical, flat or disc-shaped shell. Sponges, sea anemones, sea lilies, barnacles, oysters, sea tunicates and various corals live on the bottom of the water body and have strong reproductive capacity. Some reproduce by budding to form groups, and some produce a large number of planktonic larvae and settle down when encountering suitable substrates. Columnar worms, lampreys, rays and flounders live in the mud and sand at the bottom of the water. Clams and cockles live in holes. Some benthic organisms can be used for human consumption. Some marine and freshwater mollusks can produce pearls. It is generally believed that in water areas with soft sediments as the substrate, the density of benthic organisms decreases with increasing depth. The biomass of organisms on the continental shelf is much higher than that on the seabed plain, but in the deep sea, the species diversity is more obvious than on the continental shelf.
[0004] The current benthic organism identification has the following technical bottlenecks:
[0005] Manual identification relies on experienced experts, and there is a large gap in professional talents nationwide. Manual identification is not convenient; traditional microscope identification is time-consuming and inefficient; general identification tools have poor regional adaptability and low accuracy for special species identification; therefore, an intelligent benthic organism identification system is proposed. SUMMARY
[0006] The purpose of this section is to summarize some aspects of the embodiments of the present application and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract and title of the specification to avoid obscuring the purpose of this section, the abstract and the title. Such simplifications or omissions cannot be used to limit the scope of the present application.
[0007] In view of the above and / or existing problems in ecological environment monitoring, the present application is proposed.
[0008] Therefore, the present application aims to provide a benthic organism intelligent identification system, which deeply integrates convolutional neural network technology, combines with a million-level local database, realizes rapid and accurate identification of benthic organisms, has strong adaptive and generalization capabilities, effectively solves the industry pain points of low efficiency, unstable accuracy, poor regional adaptability and the like of traditional identification methods, provides reliable technical support for water ecological monitoring, and gradually improves the identification accuracy of unannotated biological samples through an incremental learning mechanism.
[0009] To solve the above technical problems, according to one aspect of the present application, the present application provides the following technical solutions:
[0010] A benthic organism intelligent identification system comprises:
[0011] A sample collection unit is configured to perform collection and acquisition of benthic organism microscopic images.
[0012] A feature analysis unit is connected to the sample collection unit and configured to perform multi-scale feature extraction through a convolutional neural network.
[0013] A model matching unit is connected to the feature analysis unit and configured to perform comparison with feature templates in a database.
[0014] A classification output unit is connected to the model matching unit and configured to perform output of an identification result based on probability analysis.
[0015] A data storage unit is connected to the model matching unit and configured to provide a local database.
[0016] An incremental learning unit is connected between the classification output unit and the data storage unit and configured to perform feature learning and database updating on unannotated samples.
[0017] As a preferred scheme of the benthic organism intelligent identification system, the sample collection unit adopts a high-resolution lens with three-dimensional automatic zooming capability.
[0018] As a preferred scheme of the benthic organism intelligent identification system, the feature analysis unit comprises a deep learning module, and the deep learning module adopts a convolutional neural network hierarchical structure, including a convolutional layer, a pooling layer and a fully connected layer.
[0019] As a preferred scheme of the benthic organism intelligent identification system, the model matching unit adopts a post-Bayesian confidence calculation model to generate an interpretable confidence score in the interval of 0-1.
[0020] As a preferred scheme of the benthic organism intelligent recognition system, the classification output unit synchronously performs target classification and boundary box regression based on a multi-task loss function, and outputs the final recognition result after non-maximum suppression (NMS) processing, and classifies and recognizes to the family and genus.
[0021] As a preferred scheme of the benthic organism intelligent recognition system, the data storage unit adopts a local database containing million-level species data.
[0022] As a preferred scheme of the benthic organism intelligent recognition system, the incremental learning unit adopts a self-adaptive optimization algorithm, and the model has strong robustness in a complex environment and performs incremental learning and recognition on unannotated biological samples.
[0023] Compared with the prior art, the benthic organism intelligent recognition system of the present application has the following advantages: the sample collection unit collects benthic organism microscopic images, the feature analysis unit extracts multi-scale features, then the model matching unit compares the features with the feature templates in the database, and the classification output unit outputs the recognition result based on probability analysis, and the incremental learning unit performs feature learning and database updating on unannotated biological samples. The deep fusion convolutional neural network technology is combined with the million-level local database to realize rapid and accurate recognition of benthic organisms, and the system has strong adaptive ability and generalization ability, effectively solves the industry pain points such as low efficiency, unstable accuracy, and poor regional adaptability of traditional recognition methods, and provides reliable technical support for water ecological monitoring. Through the incremental learning mechanism, the recognition accuracy of unannotated biological samples is gradually improved. BRIEF DESCRIPTION OF DRAWINGS
[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the present application will be described in detail below with reference to the drawings and detailed embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor. Among them:
[0025] Figure 1 The figure is a system structure diagram of the present application.
[0026] In the figure: 100 sample collection unit, 200 feature analysis unit, 210 deep learning module, 211 convolutional layer, 212 pooling layer, 213 fully connected layer, 300 model matching unit, 400 classification output unit, 500 data storage unit, 600 incremental learning unit. DETAILED DESCRIPTION
[0027] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings.
[0028] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to unnecessarily obscure aspects of the present application.
[0029] Secondly, the present application is described in detail in combination with the schematic diagram. In the detailed description of the embodiments of the present application, the sectional view of the device structure is partially enlarged without the general proportion for the convenience of illustration, and the schematic diagram is only an example which should not limit the scope of protection of the present application herein. In addition, the three-dimensional spatial dimensions of length, width and depth should be included in the actual manufacture.
[0030] In order to make the objectives, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below in combination with the drawings.
[0031] The present application provides a benthic organism intelligent recognition system, which deeply integrates convolutional neural network technology, combines with a million-level local database, realizes rapid and accurate recognition of benthic organisms, has strong adaptive and generalization capabilities, effectively solves the industry pain points such as low efficiency, unstable accuracy and poor regional adaptability of traditional recognition methods, provides reliable technical support for water ecological monitoring, gradually improves the recognition accuracy of unannotated biological samples through incremental learning mechanism, please refer to Figure 1 , comprising: a sample collection unit 100, a feature analysis unit 200, a model matching unit 300, a classification output unit 400, a data storage unit 500 and an incremental learning unit 600.
[0032] The sample collection unit 100 is used to execute collection and acquisition of benthic organism microscopic images;
[0033] The sample collection unit 100 adopts a high-resolution lens with three-dimensional automatic zooming capability, which can accurately focus on benthic organisms. Even if it is a small biological sample of millimeter level, its fine texture and organ structure can be clearly presented, ensuring accurate collection of morphological data. The high frame rate feature gives the sampling strong adaptability in complex water environment.
[0034] The feature analysis unit 200 is connected with the sample collection unit 100, and is used to execute extraction of multi-scale features through a convolutional neural network;
[0035] The feature analysis unit 200 comprises a deep learning module 210, which adopts a convolutional neural network hierarchical structure, including a convolutional layer 211, a pooling layer 212 and a fully connected layer 213, and adopts a convolutional neural network (CNN) as a core algorithm. The network simulates the human visual neural mechanism, and through the synergistic effect of the convolutional layer, the pooling layer and the fully connected layer, the feature information of the benthic organism image is deeply mined. In the convolutional layer, the convolution kernel scans the image through a sliding window, and captures the local features from the edge to the complex structure step by step, and finally generates a multi-dimensional feature map. The pooling layer performs down-sampling processing on the feature map, significantly reduces the data dimension and improves the operation efficiency. After multi-layer feature selection, the high-level features with strong semantic information are transmitted to the fully connected layer, and finally the accurate recognition of the biological morphology is realized.
[0036] The region-based convolutional neural network architecture takes FASTER R-CNN as the algorithm core, and realizes the positioning and recognition of benthic organisms through three-level cascade processing.
[0037] The region proposal network (RPN) generates high-recall candidate regions (ROIs) in the full image range in parallel, improving the generation efficiency.
[0038] The double-branch convolution architecture with shared weights is adopted to extract multi-level features of each ROI, and five residual modules (RESBLOCK) are used to obtain biological identification features containing texture, morphology and topological relationship.
[0039] Based on the multi-task loss function, the target classification and the boundary box regression are executed synchronously, and the final recognition result is output after the non-maximum suppression (NMS) processing.
[0040] The model matching unit 300 is connected with the feature analysis unit 200, and is used for performing comparison with the feature templates in the database.
[0041] The model matching unit 300 adopts the Bayesian confidence calculation model to generate an interpretable confidence score in the interval of 0-1.
[0042] The classification output unit 400 is connected with the model matching unit 300, and is used for executing the output of the recognition result based on the probability analysis.
[0043] The classification output unit 400 synchronously executes the target classification and the boundary box regression based on the multi-task loss function, and outputs the final recognition result after the non-maximum suppression (NMS) processing. The classification recognition is to the family and the genus.
[0044] The data storage unit 500 is connected with the model matching unit 300, and is used for executing the provision of the local database.
[0045] The data storage unit 500 adopts a local database containing data of millions of species.
[0046] The incremental learning unit 600 is connected between the classification output unit 400 and the data storage unit 500, and is used for performing feature learning and database updating on unidentified samples. The incremental learning unit 600 adopts an adaptive optimization algorithm, and the model has strong robustness in a complex environment and can perform incremental learning and recognition on unlabeled biological samples.
[0047] In specific use,
[0048] The sample collection unit 100 collects and acquires benthic biological microscopic images through a high-resolution lens, the feature analysis unit 200 extracts multi-scale features of image information through a convolutional neural network, the model matching unit 300 performs recognition and comparison on the extracted features and feature templates in the data storage unit 500, the classification output unit 400 outputs a recognition result based on probability analysis, and the incremental learning unit 600 performs feature learning and database updating on unidentified samples.
[0049] Although the present application has been described with reference to the embodiments above, various improvements can be made and components therein can be replaced with equivalents without departing from the scope of the present application. In particular, features in the embodiments disclosed in the present specification can be combined with each other in any manner as long as there is no structural conflict, and the present application is not limited to the combinations described in the present specification, which are merely provided for the purpose of omitting descriptions and saving resources. Therefore, the present application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.
Claims
1. A benthic organism intelligent recognition system, characterized in that, The application relates to a benthic organism identification system, comprising: a sample collection unit (100) for performing collection and acquisition of benthic organism microscopic images; a feature analysis unit (200) connected with the sample collection unit (100) and used for performing multi-scale feature extraction through a convolutional neural network; a model matching unit (300) connected with the feature analysis unit (200) and used for performing comparison with feature templates in a database; a classification output unit (400) connected with the model matching unit (300) and used for performing probability analysis-based output of an identification result; a data storage unit (500) connected with the model matching unit (300) and used for providing a local database; an incremental learning unit (600) connected between the classification output unit (400) and the data storage unit (500) and used for performing feature learning and database updating on un-identified samples.
2. The benthic organism intelligent identification system according to claim 1, wherein, The sample collection unit (100) adopts a high-resolution lens with three-dimensional automatic zooming capability.
3. The benthic organism intelligent identification system according to claim 1, wherein, The feature analysis unit (200) comprises a deep learning module (210) adopting a convolutional neural network hierarchical structure, and the deep learning module (210) comprises a convolutional layer (211), a pooling layer (212) and a full connection layer (213).
4. The benthic organism intelligent identification system of claim 1, wherein, The model matching unit (300) adopts a post-Bayesian confidence calculation model to generate an interpretable confidence score in the 0-1 interval.
5. The benthic organism intelligent identification system of claim 1, wherein The classification output unit (400) synchronously performs target classification and boundary box regression based on a multi-task loss function, and finally outputs an identification result after non-maximum suppression (NMS) processing, and the classification identification is performed to the family and genus.
6. The benthic organism intelligent identification system of claim 1, wherein, The data storage unit (500) adopts a local database containing million-level species data.
7. The benthic organism intelligent identification system of claim 1, wherein, The incremental learning unit (600) adopts a self-adaptive optimization algorithm, and the model has strong robustness in a complex environment and can perform incremental learning and identification on un-labeled biological samples.