A method for identifying brine shrimp echoes based on broadband acoustic multi-feature fusion
By using broadband acoustic multi-feature fusion and the SSA-XGBoost model, the problem of recognizing shrimp echoes in complex sea environments was solved, achieving high-precision and stable shrimp echo recognition.
Patent Information
- Application Number
- CN202610799020.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-04
- Publication Date
- 2026-08-25
AI Technical Summary
In existing technologies, shrimp echo identification is difficult to accurately distinguish in complex marine environments, resulting in misidentification or missed identification. Furthermore, existing methods have poor adaptability to complex mixed habitat scenarios.
A broadband acoustic multi-feature fusion method is adopted, including extracting dual-frequency difference features, time-frequency features and acoustic image texture features, and using the XGBoost model for classification. The model parameters are optimized by combining the sparrow search algorithm to form the SSA-XGBoost classification model.
It improves the accuracy and stability of krill echo recognition, effectively distinguishing krill from other small organisms in complex marine environments. It also establishes a complete technical process from raw echo data to recognition results, facilitating engineering implementation.
Smart Images
Figure CN122632233A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of marine biological acoustic detection and underwater target recognition technology, specifically, to a method for krill echo recognition that is not based on broadband acoustic multi-feature fusion. Background Technology
[0002] Krill is an important small crustacean fishery resource in my country's coastal waters, playing a crucial role in trophic level linkage within the marine ecosystem. It is also a key target for monitoring fishing quotas during fishing moratoriums, marine ecological monitoring, and biological monitoring of nuclear power plant cold source hazards. Current krill surveys primarily rely on trawls and gillnets for sampling. While these methods directly obtain samples, they suffer from problems such as sample dispersion, low operational efficiency, poor temporal continuity, insufficient spatial coverage, and disturbance to biological resources, making it difficult to meet the needs for rapid, continuous, and quantitative monitoring.
[0003] Acoustic systems can continuously acquire biological echo data over a large spatial range, offering advantages such as non-contact, speed, and high resolution. Broadband acoustic systems, in particular, can acquire wideband echoes in frequency bands such as 120 kHz and 200 kHz, providing richer spectral information for distinguishing different small biological groups. However, shrimp are small, resulting in weak target echo intensity, and their echoes are often mixed with those of juvenile fish, planktonic crustaceans, and other small marine organisms. In natural sea areas, relying solely on single-frequency intensity or dual-frequency intensity difference thresholds for discrimination can easily lead to misidentification or missed identification.
[0004] Therefore, it is necessary to establish a method for krill echo recognition that can simultaneously utilize broadband spectral differences, time-frequency structure features, and acoustic image spatial texture morphology features to improve the accuracy and stability of krill echo recognition in complex sea areas. Summary of the Invention
[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method for krill echo recognition based on broadband acoustic multi-feature fusion, in order to solve the problems of difficulty in distinguishing krill echoes from interfering organisms, insufficient recognition accuracy, and poor adaptability to complex mixed habitat scenarios in existing technologies.
[0006] To solve the above problems, the technical solution of the present invention is as follows:
[0007] A method for krill echo recognition based on broadband acoustic multi-feature fusion includes the following steps:
[0008] Collect raw broadband echo data and simultaneously sample network devices;
[0009] Preprocess the raw broadband echo data;
[0010] Extract dual-frequency difference features from candidate echo units;
[0011] Extract time-frequency features from candidate echo cells;
[0012] Candidate echo cells are mapped to acoustic image targets, and morphological and texture features are extracted.
[0013] Training of multi-feature fusion and classification models;
[0014] The multidimensional feature vector of the candidate echo unit to be identified is input into the classification model, the shrimp echo recognition result is output, and the shrimp echo recognition result is post-processed to obtain the final shrimp echo distribution result.
[0015] Preferably, in the step of acquiring raw broadband echo data and simultaneously sampling nets, a broadband split-beam scientific fish finder is used to conduct a sea survey of the target sea area. The scientific fish finder includes at least a 120 kHz transducer and a 200 kHz transducer, and uses linear frequency modulated broadband pulses for transmission and reception to acquire raw broadband echo data of different depth layers in the target sea area. The operating bandwidth of the 120 kHz transducer is 90–150 kHz, and the operating bandwidth of the 200 kHz transducer is 170–230 kHz.
[0016] Preferably, the preprocessing step of the original broadband echo data includes: system calibration based on the standard sphere method, pulse compression based on matched filtering, background noise estimation and removal, instantaneous spike noise removal and local outlier suppression, surface bubble layer and near-bottom strong reverberation region masking, and candidate echo cell extraction.
[0017] Preferably, the step of extracting dual-frequency difference features from candidate echo units specifically includes: calculating the average volumetric backscattering intensity (MVBS) at 120 kHz and 200 kHz channels for each candidate echo unit, and calculating the dual-frequency difference features. ,in, The average volumetric backscattering intensity of the transducer at 120 kHz. The average volumetric backscattering intensity of the transducer is 200 kHz.
[0018] Preferably, in the step of extracting time-frequency features from candidate echo units, the time-frequency features include one or more of the following: pulse length, echo envelope width, center frequency, effective bandwidth, spectral slope, spectral peak position, spectral energy centroid, spectral skewness, and spectral kurtosis.
[0019] Preferably, in the step of mapping candidate echo units to acoustic image targets and extracting morphological and texture features, the morphological features include one or more of area, length, width, aspect ratio, perimeter, compactness, boundary complexity, and circularity; the texture features include one or more of contrast, energy, homogeneity, correlation, and entropy.
[0020] Preferably, in the process of multi-feature fusion and classification model training, the classification model is an XGBoost model, and the parameters of the XGBoost model are optimized using the Sparrow Search Algorithm (SSA) to form an SSA-XGBoost classification model.
[0021] Preferably, the multidimensional feature vector of the candidate echo unit to be identified is input into the trained SSA-XGBoost classification model, and the class label of the candidate echo unit or the probability value of belonging to the shrimp is output.
[0022] Preferably, the post-processing of the shrimp echo recognition results specifically includes one or more of the following: neighborhood consistency check, small connected component deletion, track direction smoothing, adjacent depth layer consistency correction, and isolated target removal.
[0023] Compared with the prior art, the present invention has the following beneficial effects:
[0024] 1. This invention makes full use of broadband continuous spectrum information to improve the separability of shrimp echoes from other small organisms;
[0025] 2. This invention integrates dual-frequency difference features, time-frequency features, and image morphological texture features to overcome the problem of unstable recognition of single features;
[0026] 3. This invention introduces SSA to optimize the XGBoost model, which improves the adaptability and robustness of the classifier;
[0027] 4. This invention establishes a complete technical process from raw echo data to shrimp recognition results, which facilitates engineering implementation. Attached Figure Description
[0028] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:
[0029] Figure 1 This is a flowchart of the shrimp echo recognition method based on broadband acoustic multi-feature fusion of the present invention.
[0030] Figure 2 This is a schematic diagram of the echo recognition results for krill. Detailed Implementation
[0031] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.
[0032] Specifically, this invention provides a method for krill echo recognition based on broadband acoustic multi-feature fusion, such as... Figure 1 and Figure 2 As shown, the method includes the following steps:
[0033] S1: Collect raw broadband echo data and simultaneously sample network devices;
[0034] Specifically, a broadband split-beam scientific fish finder is used to conduct a sea survey of the target sea area. The scientific fish finder includes at least a 120 kHz transducer and a 200 kHz transducer, and uses linear frequency modulated broadband pulses for transmission and reception to acquire raw broadband echo data of different depth layers in the target sea area.
[0035] Simultaneously, biological sampling was conducted using nets or trawls to obtain samples of krill and major interfering organisms, and the sampling station location, time, latitude and longitude, water depth, sample species composition, and body length distribution information were recorded.
[0036] Preferably, the 120 kHz channel has a working bandwidth of 90–150 kHz, the 200 kHz channel has a working bandwidth of 170–230 kHz, the transmit pulse duration is 0.512 ms, 1.024 ms or 2.048 ms, preferably 1.024 ms; and the travel speed is 6–8 knots, preferably 7 knots.
[0037] S2: Preprocess the raw broadband echo data;
[0038] Specifically, the raw broadband echo data obtained in step S1 is preprocessed, including: system calibration based on the standard sphere method, pulse compression based on matched filtering, background noise estimation and removal, instantaneous spike noise removal and local outlier suppression, masking of the surface bubble layer and near-bottom strong reverberation region, and extraction of candidate echo cells.
[0039] S3: Extract dual-frequency difference features from candidate echo units;
[0040] Specifically, for the candidate echo cells obtained in step S2, the average volumetric backscattering intensity (MVBS) at 120 kHz and 200 kHz channels is calculated, and the dual-frequency difference characteristics are also calculated.
[0041]
[0042] in, The average volumetric backscattering intensity of the transducer at 120 kHz. The average volumetric backscattering intensity of the transducer is 200kHz. Further, frequency difference-related features such as average band difference, band energy ratio, and local peak-valley difference can be extracted. Preferably, the target is a krill. The prior distribution range was set to -6 dB to 2 dB, and this range was used as one of the prior screening conditions for subsequent classification.
[0043] S4: Extract time-frequency features from candidate echo units;
[0044] Specifically, the time-frequency characteristics include one or more of the following: pulse length, echo envelope width, center frequency, effective bandwidth, spectral slope, spectral peak position, spectral energy centroid, spectral skewness, and spectral kurtosis. Preferably, the above parameters are obtained by short-time Fourier transform (STFT) or matched-filter output spectrum analysis.
[0045] S5: Map candidate echo units to acoustic image targets and extract morphological and texture features;
[0046] Specifically, the morphological features include one or more of the following: area, length, width, aspect ratio, perimeter, compactness, boundary complexity, and roundness.
[0047] The texture features are extracted using a Gray-Level Co-occurrence Matrix (GLCM), with the following preferred parameters: 16 gray-level quantization levels; window size of 7×7 or 9×9, preferably 7×7; orientation angles of 0°, 45°, 90°, and 135°; and pixel pitch of 1. The texture features extracted from the GLCM include one or more of the following: contrast, energy, homogeneity, correlation, and entropy.
[0048] S6: Perform multi-feature fusion and classification model training;
[0049] Specifically, all features obtained in steps S3, S4 and S5 are standardized to construct a multidimensional feature vector for candidate echo units.
[0050] Training and testing sets were established based on tag samples obtained from synchronous net sampling and manual verification.
[0051] A classification model was constructed using XGBoost, and the parameters of the XGBoost model were optimized using the Sparrow Search Algorithm (SSA) to form the SSA-XGBoost classification model.
[0052] S7: Input the multidimensional feature vector of the candidate echo unit to be identified into the classification model, output the shrimp echo recognition result, and perform post-processing on the shrimp echo recognition result to obtain the final shrimp echo distribution result.
[0053] Specifically, the feature vector of the candidate echo unit to be identified is input into the trained SSA-XGBoost classification model, and the class label of the candidate echo unit or the probability value of belonging to the shrimp is output.
[0054] The shrimp echo identification results are post-processed to obtain the final shrimp echo distribution results. The post-processing includes one or more of the following: neighborhood consistency check, small connected component deletion, track direction smoothing, adjacent depth layer consistency correction, and isolated target removal.
[0055] The following specific embodiments illustrate the shrimp echo recognition method based on broadband acoustic multi-feature fusion of the present invention.
[0056] A broadband acoustic system was used to conduct a sea survey in the Haizhou Bay area. The 120 kHz transducer had a bandwidth of 90–150 kHz, and the 200 kHz transducer had a bandwidth of 170–230 kHz. The transmission pulse duration was set to 1.024 ms, and the survey speed was approximately 7 knots. Simultaneously, small trawls were used to collect samples at typical stations to obtain information on the species composition and body length distribution of shrimp, larvae, and other crustaceans.
[0057] After the raw acoustic data is imported into the processing system, standard sphere calibration and matched filter pulse compression are performed first, followed by background noise estimation and removal, spike noise removal, and local median filtering. After masking the 2 m surface area and the 1 m near bottom area, candidate echo cells are extracted with a minimum echo threshold of -85 dB and a minimum connected region condition of 8 pixels.
[0058] For each candidate echo unit, the MVBS at 120 kHz and 200 kHz is calculated, and the ΔMVBS is obtained. At the same time, time-frequency features such as pulse length, spectral slope, spectral peak position, and spectral energy centroid are extracted. Furthermore, features such as area, aspect ratio, boundary complexity, GLCM contrast, energy, and entropy are extracted from the acoustic image to form a multi-dimensional feature vector.
[0059] Based on the annotation of krill and non-krill samples from synchronous trawl samples, training and test sets were established. SSA was used to optimize XGBoost parameters for model training, and a classification model was obtained through five-fold cross-validation. The model was used to identify all candidate echo units, and the krill echo distribution results were output after processing with neighborhood consistency and smoothing rules. The results show that this invention has high recognition accuracy and strong stability for krill echoes in complex mixed habitat backgrounds.
[0060] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.
Claims
1. A method for shrimp echo recognition based on broadband acoustic multi-feature fusion, characterized in that, The method includes: Collect raw broadband echo data and simultaneously sample network devices; Preprocess the raw broadband echo data; Extract dual-frequency difference features from candidate echo units; Extract time-frequency features from candidate echo cells; Candidate echo cells are mapped to acoustic image targets, and morphological and texture features are extracted. Training of multi-feature fusion and classification models; The multidimensional feature vector of the candidate echo unit to be identified is input into the classification model, the shrimp echo recognition result is output, and the shrimp echo recognition result is post-processed to obtain the final shrimp echo distribution result.
2. The method for shrimp echo recognition based on broadband acoustic multi-feature fusion according to claim 1, characterized in that, In the step of acquiring raw broadband echo data and simultaneously sampling nets, a broadband split-beam scientific fish finder is used to conduct a sea survey of the target sea area. The scientific fish finder includes at least a 120 kHz transducer and a 200 kHz transducer, and uses linear frequency modulated broadband pulses for transmission and reception to acquire raw broadband echo data of different depth layers in the target sea area. The operating bandwidth of the 120 kHz transducer is 90–150 kHz, and the operating bandwidth of the 200 kHz transducer is 170–230 kHz.
3. The method for shrimp echo recognition based on broadband acoustic multi-feature fusion according to claim 1, characterized in that, The preprocessing steps for the raw broadband echo data include: system calibration based on the standard sphere method, pulse compression based on matched filtering, background noise estimation and removal, instantaneous spike noise removal and local outlier suppression, surface bubble layer and near-bottom strong reverberation region masking, and candidate echo cell extraction.
4. The method for shrimp echo recognition based on broadband acoustic multi-feature fusion according to claim 1, characterized in that, The step of extracting dual-frequency difference features from candidate echo units specifically includes: calculating the average volumetric backscattering intensity (MVBS) at 120 kHz and 200 kHz channels for each candidate echo unit, and calculating the dual-frequency difference features. ,in, The average volumetric backscattering intensity of the transducer at 120 kHz. The average volumetric backscattering intensity of the transducer is 200 kHz.
5. The method for shrimp echo recognition based on broadband acoustic multi-feature fusion according to claim 1, characterized in that, In the step of extracting time-frequency features from candidate echo units, the time-frequency features include one or more of the following: pulse length, echo envelope width, center frequency, effective bandwidth, spectral slope, spectral peak position, spectral energy centroid, spectral skewness, and spectral kurtosis.
6. The method for shrimp echo recognition based on broadband acoustic multi-feature fusion according to claim 1, characterized in that, In the step of mapping candidate echo units to acoustic image targets and extracting morphological and texture features, the morphological features include one or more of area, length, width, aspect ratio, perimeter, compactness, boundary complexity, and circularity; the texture features include one or more of contrast, energy, homogeneity, correlation, and entropy.
7. The method for shrimp echo recognition based on broadband acoustic multi-feature fusion according to claim 1, characterized in that, In the process of multi-feature fusion and classification model training, the classification model is an XGBoost model, and the parameters of the XGBoost model are optimized using the Sparrow Search Algorithm (SSA) to form an SSA-XGBoost classification model.
8. The method for shrimp echo recognition based on broadband acoustic multi-feature fusion according to claim 7, characterized in that, The multidimensional feature vector of the candidate echo unit to be identified is input into the trained SSA-XGBoost classification model, which outputs the category label of the candidate echo unit or the probability value of it belonging to a shrimp.
9. The method for shrimp echo recognition based on broadband acoustic multi-feature fusion according to claim 1, characterized in that, The post-processing of the shrimp echo recognition results specifically includes one or more of the following: neighborhood consistency check, small connected component deletion, track direction smoothing, adjacent depth layer consistency correction, and isolated target removal.