Fish school density monitoring method based on big data and sound wave signals

Abstract

The invention discloses a fish school density monitoring method based on big data and sound wave signals, and the method comprises the steps: constructing a sound wave reflection source type recognition model, and determining the type and position of a sound wave reflection source of a sound wave reflection signal; according to the fish school sound wave reflection signals, multi-path reflection signals in the same sound wave reflection source are recognized and removed; constructing a fish school density prediction model, predicting fish school density, and determining a fish school activity index; according to the reference fish school activity index threshold value and the fish school activity index, the fish school health state is judged; according to the fish body appearance image with the abnormal health state, the fish school disease category is judged; according to the fish school disease types, fish school health intervention measures are made, and the effectiveness of the fish school health intervention measures is judged. The real-time performance and precision of fish school density and fish school health monitoring are improved, the risk in aquaculture is reduced, and the efficiency and scientificity of aquaculture management are greatly improved through automatic water quality regulation and control and chemical dosing equipment.

Description

Technical Field

[0001] The present invention relates to the field of aquaculture monitoring, and in particular to a fish density monitoring method based on big data and acoustic wave signals. Background Art

[0002] In the aquaculture industry, real-time monitoring of fish density, health status and growth environment is crucial to improving aquaculture benefits and ensuring the health of fish. However, current monitoring methods have many limitations, resulting in low efficiency and high cost of aquaculture management, especially when fish health problems are often discovered late, missing the best time for intervention. This management deficiency not only affects the benefits of aquaculture, but may also lead to risks such as large-scale mortality of fish and environmental deterioration. Therefore, how to improve the accuracy and real-time nature of monitoring has become a difficult problem that the industry needs to solve urgently. In order to solve these problems, acoustic wave detection technology has gradually become an important means of monitoring fish. This technology relies on the propagation and reflection principle of sound waves in water bodies. By analyzing the reflected signals after the sound waves encounter the fish, the location and density of the fish are determined. Compared with traditional manual detection methods, acoustic wave detection technology can provide more real-time information on the distribution of fish schools and reduce the delay of relying on human detection. However, although acoustic wave detection technology has made great progress in improving monitoring efficiency, there are still many technical challenges in practical applications. For example, the underwater environment is complex and changeable, and there are factors such as water flow, temperature, and salinity that make the acoustic wave signal easily interfered. Especially in larger waters, multipath reflection of sound waves is more common, which will cause repetition and confusion of reflected signals and affect the accuracy of detection. Multipath reflection signals refer to multiple reflection signals generated when sound waves encounter obstacles or different layers of water bodies. These signals are superimposed together, which can easily lead to misjudgment of the position and density of fish schools, thereby affecting the reliability of data. In addition to the multipath reflection problem, most existing sound wave detection equipment can only provide basic information on the distribution of fish schools, and it is difficult to capture more complex dynamic characteristics of fish schools, such as the health status, swimming speed and behavioral changes of fish schools. Most existing technologies can only monitor the static density of fish schools, and cannot effectively track the real-time movement trend and activity index of fish schools, resulting in the inability to provide timely warnings in the early stages of abnormal behavior or health problems of fish schools. In addition, the processing accuracy and resolution of sound wave detection signals are also limited by equipment and algorithms, making it difficult to accurately predict changes in the density of fish schools. For large-scale and intensive aquaculture, this technical deficiency will lead to unscientific density management, which in turn affects the health of fish schools and breeding efficiency. In addition, the health management of fish schools mainly relies on manual detection and regular sampling, which is not only time-consuming and labor-intensive, but also has a large delay. When fish have health problems, external symptoms are usually difficult to detect in the early stages. By the time the disease spreads, the best time for intervention has often been missed. As the scale of aquaculture expands, the limitations of manual detection methods become more and more obvious, and fish health monitoring becomes more difficult, making it difficult for farmers to take effective measures in time to prevent the spread of diseases. Although acoustic wave detection technology has been widely used in fish monitoring, there are still problems such as multipath reflection signal interference, low accuracy of monitoring data, lack of real-time performance, and lagging health management.These problems seriously affect the decision-making efficiency and risk control in the breeding process. Therefore, a new technical means is urgently needed to improve the accuracy, real-time and automation level of fish monitoring to better support scientific breeding and environmental management. Summary of the invention

[0003] The present invention solves the problems existing in the above-mentioned prior art and provides a fish density monitoring method based on big data and acoustic wave signals, which mainly includes:

[0004] According to the sound wave reflection signal of the water area to be detected, a sound wave reflection source type identification model is constructed to determine the type of the sound wave reflection source of the sound wave reflection signal and the position of the sound wave reflection source;

[0005] According to the acoustic wave reflection signals of the school of fish, the similarity of the frequency spectrum characteristics of the acoustic wave reflection signals of the school of fish is calculated, and the multi-path reflection signals in the same acoustic wave reflection source are identified and eliminated;

[0006] Based on the echo width, spectrum characteristics, phase difference and phase offset of the fish school acoustic reflection signal, a fish school density prediction model is constructed to predict the fish school density, and the fish school activity index is determined based on the swimming speed, density, swimming speed fluctuation and speed change rate of the fish school in the water area;

[0007] The fish health status is determined based on the baseline fish activity index threshold and the fish activity index, and the drone-mounted camera is used to obtain fish body appearance images of fish with abnormal health status;

[0008] Based on the appearance images of fish with abnormal health status, identify the behavioral characteristics and external symptoms of the fish, and determine the type of fish disease;

[0009] According to the types of fish diseases, fish health intervention measures are formulated and sent to automated water quality control equipment and automatic drug delivery equipment for fish health management, real-time monitoring of fish activity index, and judgment of the effectiveness of fish health intervention measures.

[0010] Furthermore, the method of constructing a sound wave reflection source type identification model according to the sound wave reflection signal of the water area to be detected, determining the sound wave reflection source type of the sound wave reflection signal, and determining the position of the sound wave reflection source includes:

[0011] The acoustic detection equipment emits acoustic waves to the water area to be detected according to the set acoustic emission parameters, and the acoustic reflection signal and the corresponding timestamp information are obtained through the directional array sensor; according to the acoustic reflection signal of the water area to be detected, the acoustic reflection signal is separated by the independent component analysis algorithm to obtain the acoustic reflection signals of different acoustic reflection sources, and saved in the acoustic reflection monitoring database; the time domain acoustic reflection signal is converted into frequency domain data by the short-time Fourier transform algorithm, and its spectrum characteristics are extracted, and the spectrum characteristics include frequency, frequency peak and amplitude; through the acoustic reflection monitoring database, the historical acoustic emission frequency, amplitude, and spectrum characteristics of the acoustic reflection signal are obtained, and the type of acoustic reflection source is marked, and the support vector machine algorithm is used for model training to build an acoustic reflection source type recognition model to determine the acoustic reflection source type of the acoustic reflection signal, including fish school acoustic reflection source and non-fish school acoustic reflection source; the echo time of the acoustic reflection signal of the fish school is obtained, and the position of the fish school relative to the sensor array is calculated by time difference estimation to determine the position of the fish school.

[0012] Further, the similarity of the frequency spectrum characteristics of the fish school sound wave reflection signals is calculated according to the fish school sound wave reflection signals, and the multipath reflection signals in the same sound wave reflection source are identified and eliminated, including:

[0013] The acoustic wave reflection signals of the fish school are obtained through the acoustic wave reflection monitoring database, and the similarity of the spectral characteristics of the acoustic wave reflection signals of the fish school is calculated using the method of calculating the Pearson correlation coefficient. If the similarity is greater than a preset similarity threshold and the echo time difference is less than a preset difference threshold, the acoustic wave reflection signal of the fish school is determined to be a multi-path reflection signal of the same acoustic wave reflection source; the multi-path reflection signals in the same acoustic wave reflection source are eliminated, and the unique acoustic wave reflection signal of the acoustic wave reflection source is retained.

[0014] Further, the fish school density prediction model is constructed based on the echo width, spectrum characteristics, phase difference and phase offset of the fish school acoustic wave reflection signal to predict the fish school density, and the fish school activity index is determined according to the swimming speed, density, swimming speed volatility and speed change rate of the fish school in the water area, including:

[0015] According to the unique sound wave reflection signal of the sound wave reflection source, the phase difference and phase offset of the sound wave transmission signal are determined; based on the echo width, spectral characteristics, phase difference and phase offset of the sound wave reflection signal of the fish school, a recurrent neural network algorithm is used for model training to build a fish school density prediction model to predict the fish school density; the fish school density is continuously monitored. If the fish school density change rate is higher than the preset density threshold, the camera mounted on the drone is used to obtain the fish body appearance image of the fish school and save it to the fish school appearance image database; if the fish school density change rate is lower than the preset density threshold, the echo signals at different time points are compared to judge the position change of the fish school and determine the swimming speed of the fish school; the fish school activity index is determined according to the swimming speed, density, swimming speed volatility and speed change rate of the fish school in the water area.

[0016] It also includes determining the fish activity index based on the swimming speed, density, swimming speed fluctuation and speed change rate of the fish school in the water area, including:

[0017] According to the swimming speed of the fish school in the water area, the fish school swimming speed fluctuation formula is used Calculate the fluctuation of fish school swimming speed σ V (t), where V(t) is the swimming speed of the fish school at time t, is the average swimming speed in the time period T; according to the swimming speed, density, swimming speed fluctuation and speed change rate of the fish school in the water area, the fish school activity index formula is used Calculate the fish activity index FAI(t), where V(t) is the average swimming speed of the fish, D(t) is the fish density, and σ V (t) is the fluctuation of the swimming speed of the fish school, is the rate of change of the swimming speed of the fish school.

[0018] Further, judging the health status of the fish school according to the reference fish school activity index threshold and the fish school activity index, and obtaining the fish body appearance image of the fish school with abnormal health status by using the camera carried by the drone, includes:

[0019] According to the calculated fish activity index, the water environment parameters of the fish activity index corresponding to the detection time are obtained in real time through the water environment sensor and saved in the fish monitoring database. The water environment parameters include flow rate, dissolved oxygen content, water temperature, and turbidity; according to the water environment parameters obtained in real time, the baseline fish activity index threshold is adjusted; according to the baseline fish activity index threshold and the fish activity index, the health status of the fish is judged, and the health status of the fish is normal and abnormal; if the difference between the fish activity index and the baseline fish activity index threshold is greater than the preset difference threshold, the health status of the fish is judged to be abnormal, and the camera mounted on the drone is used to obtain the fish appearance image of the fish with abnormal health status, and save it to the fish appearance image database.

[0020] It also includes adjusting the threshold of the baseline fish activity index based on the water environment parameters obtained in real time, including:

[0021] Through the fish monitoring database, historical water environment parameters are obtained, and the benchmark fish activity index thresholds corresponding to the historical water environment parameters are marked. The random forest regression algorithm is used for model training to build a fish activity index threshold adjustment model. According to the real-time water environment parameters, the fish activity index threshold adjustment model is used to adjust the benchmark fish activity index threshold. The health status of the fish is judged based on the adjusted benchmark fish activity index threshold and the real-time fish activity index.

[0022] Further, the identifying of fish behavior characteristics and external symptoms of fish based on the fish appearance images with abnormal health status, and determining the type of fish disease, includes:

[0023] Through the fish appearance image database, historical fish appearance images with abnormal health status are obtained, and behavioral characteristics and external symptoms are marked. The convolutional neural network is used for model training to build a fish feature recognition model; based on the real-time acquired fish appearance images with abnormal health status and the fish appearance images of fish with a density change rate higher than a preset density threshold, the fish behavior characteristics and fish external symptoms are identified. The fish behavior characteristics include but are not limited to abnormal swimming and floating heads, and the fish external symptoms include but are not limited to surface lesions, decolorization, abnormal scales, ulcers, and parasites; the fish disease category is determined based on the fish behavior characteristics and fish external symptoms; and the fish health status and fish disease category are sent to the fish health management center in real time through fish health monitoring equipment.

[0024] Furthermore, according to the fish disease category, fish health intervention measures are formulated and sent to the automated water quality control equipment and the automatic drug delivery equipment for fish health management, real-time monitoring of fish activity index, and determination of the effectiveness of fish health intervention measures, including:

[0025] Through the dispatching system of the fish health management center, fish health intervention measures are formulated according to the types of fish diseases. The intervention measures include water quality adjustment requirements, required drug types and dosages, and the intervention measures are recorded and sent to the automated water quality control equipment and automatic drug delivery equipment; oxygen aeration equipment and pH value adjustment equipment are used to adjust water quality parameters, and automatic drug delivery equipment is used to inject drugs into the water body according to the preset delivery rate and cycle according to the disease prevention or treatment requirements in the fish health adjustment work order, and the drug delivery information is transmitted to the fish health management center; the fish activity index is monitored in real time, and the current fish activity index is compared with the benchmark fish activity index threshold to determine whether the fish health intervention measures are effective, and the feedback data is sent to the fish health management center; if the fish health intervention measures are effective and the fish health status returns to normal, the water quality control equipment and automatic drug delivery equipment are shut down through the dispatching system of the fish health management center, and the water quality adjustment and drug delivery are stopped; if the fish health intervention measures are ineffective, new fish health intervention measures are generated, and the water quality control equipment and automatic drug delivery equipment are readjusted until the fish health status returns to normal.

[0026] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0027] The present invention provides a fish density monitoring method based on big data and acoustic wave signals. By constructing an acoustic wave reflection source identification model and a fish density prediction model, the present invention can effectively identify and eliminate multipath reflection signals, thereby improving the accuracy of fish position and density monitoring. The present invention calculates the fish activity index by analyzing the swimming speed, density, swimming speed volatility and speed change rate of the fish, and can obtain the behavioral dynamics and health status of the fish in real time. The present invention can also dynamically adjust the baseline activity index threshold according to the water environment parameters to ensure the accuracy of the assessment under different water quality conditions. When the health of the fish is abnormal, the appearance image of the abnormal fish is automatically collected and analyzed by a drone, and the type of fish disease is quickly determined and health intervention measures are provided. The present invention not only improves the real-time and accuracy of fish density and fish health monitoring, reduces the risks in aquaculture, but also greatly improves the efficiency and scientificity of aquaculture management through automated water quality control and drug administration equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 A flow chart of a fish density monitoring method based on big data and acoustic wave signals of the present invention;

[0029] Figure 2 A schematic diagram of a fish density monitoring method based on big data and acoustic wave signals of the present invention;

[0030] Figure 3It is another schematic diagram of a fish density monitoring method based on big data and acoustic wave signals of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1-3 In this embodiment, a fish density monitoring method based on big data and acoustic wave signals may specifically include:

[0033] Step S101, constructing a sound wave reflection source type recognition model according to the sound wave reflection signal of the water area to be detected, determining the sound wave reflection source type of the sound wave reflection signal, and determining the position of the sound wave reflection source.

[0034] The acoustic detection equipment emits acoustic waves to the waters to be detected according to the set acoustic emission parameters, and the acoustic reflection signal and the corresponding timestamp information are obtained through the directional array sensor. According to the acoustic reflection signal of the waters to be detected, the acoustic reflection signal is separated by the independent component analysis algorithm to obtain the acoustic reflection signals of different acoustic reflection sources, and saved in the acoustic reflection monitoring database. The time domain acoustic reflection signal is converted into frequency domain data using the short-time Fourier transform algorithm, and its spectrum characteristics are extracted. The spectrum characteristics include frequency, frequency peak and amplitude. Through the acoustic reflection monitoring database, the historical acoustic emission frequency, amplitude, and spectrum characteristics of the acoustic reflection signal are obtained, and the acoustic reflection source type is marked. The support vector machine algorithm is used for model training to build an acoustic reflection source type recognition model to determine the acoustic reflection source type of the acoustic reflection signal, including fish school acoustic reflection source and non-fish school acoustic reflection source. The echo time of the acoustic reflection signal of the fish school is obtained, and the position of the fish school relative to the sensor array is calculated by time difference estimation to determine the position of the fish school.

[0035] Exemplarily, in an aquaculture farm, the acoustic wave detection equipment is set to emit acoustic wave signals at a frequency of 80kHz. These acoustic waves propagate in the water and are received by directional array sensors arranged at different positions, and the timestamp information of each acoustic wave reflection signal is recorded by the sensor. The acoustic wave reflection signal is processed by an independent component analysis algorithm to separate the signals from different sound sources. After processing, the acoustic wave reflection signals from two main sound sources are obtained, one from the school of fish and the other from the facilities or environmental structures in the water. The time domain acoustic wave reflection signal is converted into frequency domain data using a short-time Fourier transform algorithm. After processing, the following spectral characteristics are obtained, including that the frequency of the acoustic wave reflection signal of the school of fish is 75kHz, the frequency peak is at 80kHz, and the amplitude is 0.2 volts. Through the acoustic wave reflection monitoring database, the acoustic wave emission frequency and amplitude data of the historical records, as well as the spectral characteristics of other acoustic wave reflection signals, are obtained, and the support vector machine algorithm is used to train the spectral feature data model, and a sound wave reflection source type recognition model is constructed to identify whether the acoustic wave reflection signal comes from the school of fish or other objects in the water. By recording the echo time of the sound wave reflection signal, the position of the fish school relative to the sensor array can be estimated based on the time difference. If the echo time of the sound wave reflection is 0.6 seconds, and the speed of sound wave propagation in water is 1500 meters per second, then the position of the fish school can be calculated to be about 900 meters away from the sensor.

[0036] Step S102, based on the sound wave reflection signals of the school of fish, the similarity of the frequency spectrum characteristics of the sound wave reflection signals of the school of fish is calculated, and the multi-path reflection signals in the same sound wave reflection source are identified and eliminated.

[0037] The acoustic reflection signals of the fish school are obtained through the acoustic reflection monitoring database, and the similarity of the spectral characteristics of the acoustic reflection signals of the fish school is calculated by calculating the Pearson correlation coefficient. If the similarity is greater than the preset similarity threshold and the echo time difference is less than the preset difference threshold, the acoustic reflection signal of the fish school is determined to be a multi-path reflection signal of the same acoustic reflection source. The multi-path reflection signals in the same acoustic reflection source are eliminated, and the unique acoustic reflection signal of the acoustic reflection source is retained.

[0038] Exemplarily, the acoustic reflection signal data of two schools of fish, signal A and signal B, are obtained through the acoustic reflection monitoring database. Signal A and signal B are converted from the time domain to frequency domain data using short-time Fourier transform, and the spectral characteristics of signal A and signal B are extracted, including frequency, frequency peak and amplitude. If the frequency of signal A is 75kHz, the frequency peak is at 80kHz, the amplitude is 0.3F, and the frequency of signal B is 76kHz, the frequency peak is at 81kHz, and the amplitude is 0.29F. The similarity between the spectral characteristics of signal A and signal B is calculated by calculating the Pearson correlation coefficient calculation method. If the calculated Pearson correlation coefficient is 0.95, it is greater than the preset small similarity threshold of 0.9. At the same time, check the echo time difference between signal A and signal B. If the echo time of signal A is 0.6 seconds and the echo time of signal B is 0.59 seconds. According to the preset difference threshold of 0.05, if the echo time difference is less than 0.1 seconds, it is determined that the echo time difference between signal A and signal B is within the allowable range. The similarity of the spectrum characteristics of signal A and signal B is higher than the preset similarity threshold of 0.9, and the echo time difference between signal A and signal B is less than the preset difference threshold of 0.05 seconds. Therefore, it is determined that signal A and signal B are multipath reflection signals from the same sound wave reflection source.

[0039] Step S103, based on the echo width, spectrum characteristics, phase difference and phase offset of the fish school acoustic reflection signal, a fish school density prediction model is constructed to predict the fish school density, and the fish school activity index is determined according to the swimming speed, density, swimming speed volatility and speed change rate of the fish school in the water area.

[0040] According to the unique sound wave reflection signal of the sound wave reflection source, the phase difference and phase offset of the sound wave transmission signal are determined. Based on the echo width, spectral characteristics, phase difference and phase offset of the sound wave reflection signal of the fish school, the recurrent neural network algorithm is used for model training to build a fish school density prediction model to predict the fish school density. The fish school density is continuously monitored. If the fish school density change rate is higher than the preset density threshold, the camera mounted on the drone is used to obtain the fish body appearance image of the fish school and save it to the fish school appearance image database. If the fish school density change rate is lower than the preset density threshold, the echo signals at different time points are compared to determine the position change of the fish school and the swimming speed of the fish school. The fish school activity index is determined based on the swimming speed, density, swimming speed volatility and speed change rate of the fish school in the water area.

[0041] For example, in an aquaculture farm, the acoustic detection equipment transmits an acoustic signal with a frequency of 200kHz, and the echo width of the received acoustic reflection signal of the fish school is 2 milliseconds. The spectrum characteristics show that the peak frequency is 195kHz, the amplitude is 0.8 units, the phase difference is 15 degrees, and the phase shift is 5 degrees. Based on these data, the recurrent neural network algorithm is used for model training to build a fish density prediction model, and the fish density in the area is predicted to be 50 / cubic meter. The fish density is continuously monitored. If the fish density change rate is higher than the preset density threshold of 10 / cubic meter, the drone-mounted camera is used to obtain the fish body appearance image of the fish school and save it to the fish appearance image database. If the fish density change rate is lower than the preset density threshold of 10 per cubic meter, by comparing the position change of the fish reflection signal at different time points, such as every 5 seconds, the swimming speed of the fish is determined to be 0.2 meters per second, and the fluctuation of the swimming speed of the fish in the water area is 0.03 meters per second, and the speed change rate is 0.01 meters per second2. Combining these parameters, the activity index of the fish is calculated to be 0.75.

[0042] Among them, the fish activity index is determined based on the swimming speed, density, swimming speed volatility and speed change rate of the fish in the water area.

[0043] According to the swimming speed of the fish school in the water area, the fish school swimming speed fluctuation formula is used Calculate the fluctuation of fish school swimming speed σ V (t), where V(t) is the swimming speed of the fish school at time t, It is the average swimming speed in the time period T. According to the swimming speed, density, swimming speed fluctuation and speed change rate of the fish school in the water area, the fish school activity index formula is used Calculate the fish activity index FAI(t), where V(t) is the average swimming speed of the fish, D(t) is the fish density, and σ V (t) is the fluctuation of the swimming speed of the fish school, is the rate of change of the swimming speed of the fish school.

[0044] For example, in a water area of ​​an aquaculture farm, the swimming data of a school of fish is recorded using an acoustic detection system. At a certain time t, the instantaneous swimming speed V(t) of the school of fish is 1.5 m / s, and the density of the school of fish D(t) is 8 fish per cubic meter. By analyzing the speed data of the last 10 seconds, the average swimming speed of the school of fish is calculated. is 1.3 m / s. According to the formula of the fluctuation of the swimming speed of fish schools Calculate the fluctuation of fish school swimming speed σ V (t) is 0.25 m / s, reflecting the variation of the swimming speed of the fish school during this period. The greater the fluctuation of the swimming speed of the fish school, the stronger the activity of the fish school. At the same time, the speed change rate of the fish school is recorded. The fish activity index formula is 0.1 m / s2, and is calculated based on the swimming speed, density, swimming speed fluctuation and speed change rate of the fish in the water area. The calculated fish activity index FAI(t) is 6.8. The higher the value of the fish activity index, the more frequent the fish activity, which reflects the health and activity of the fish in the waters.

[0045] Step S104, judging the health status of the fish school according to the benchmark fish school activity index threshold and the fish school activity index, and using the camera mounted on the drone to obtain the appearance image of the fish body of the fish school with abnormal health status.

[0046] According to the calculated fish activity index, the water environment parameters corresponding to the detection time of the fish activity index are obtained in real time through the water environment sensor and saved in the fish monitoring database. The water environment parameters include flow rate, dissolved oxygen content, water temperature, and turbidity. According to the water environment parameters obtained in real time, the baseline fish activity index threshold is adjusted. According to the baseline fish activity index threshold and the fish activity index, the health status of the fish is judged, and the health status of the fish is normal and abnormal. If the difference between the fish activity index and the baseline fish activity index threshold is greater than the preset difference threshold, the health status of the fish is judged to be abnormal, and the appearance image of the fish body of the fish with abnormal health status is obtained by using the camera carried by the drone, and saved in the fish appearance image database.

[0047] Exemplarily, according to the calculated fish activity index, the environmental parameters of the water body corresponding to the detection time of the fish activity index are obtained in real time through the water environment sensor, and these data are saved in the fish monitoring database. If at the moment of monitoring, the flow rate is 0.3 m / s, the dissolved oxygen content is 8 mg / L, the water temperature is 22 degrees Celsius, and the turbidity is 5 NTU. If under the current environmental conditions, the baseline fish activity index threshold is set to 7.0. Compare the current fish activity index FAI (t) = 6.8 with the baseline threshold of 7.0, the difference is 0.2, which is lower than the preset difference threshold of 1.0. Therefore, it can be judged that the health status of the fish school is normal. If the difference between the monitored fish activity index and the baseline threshold exceeds the difference threshold of 1.0, such as the fish activity index reaches 8.2, the difference is 1.2 at this time, which is greater than 1.0, then the fish health status is automatically judged to be abnormal. If the health status of the fish school is determined to be abnormal, the drone-mounted camera will be used to capture the fish body appearance image of the abnormal fish school, and the fish body appearance image of the fish school will be uploaded and saved in the fish school appearance image database.

[0048] Among them, the benchmark fish activity index threshold is adjusted according to the water environment parameters obtained in real time.

[0049] Through the fish monitoring database, historical water environment parameters are obtained, and the baseline fish activity index threshold corresponding to the historical water environment parameters is marked. The random forest regression algorithm is used for model training to build a fish activity index threshold adjustment model. According to the real-time water environment parameters, the fish activity index threshold adjustment model is used to adjust the baseline fish activity index threshold. The health status of the fish is judged based on the adjusted baseline fish activity index threshold and the real-time fish activity index.

[0050] For example, in an aquaculture farm, a large amount of historical water environment data and corresponding fish activity conditions are collected through the monitoring database. The water environment data includes environmental parameters such as flow rate, dissolved oxygen content, water temperature and turbidity. For example, in a certain historical period, the water temperature is 20 degrees Celsius, the flow rate is 0.4 meters per second, the dissolved oxygen content is 9 milligrams per liter, and the turbidity is 6NTU. The corresponding benchmark fish activity index threshold is 6.5. The historical water environment data is annotated with the benchmark fish activity index threshold, and the random forest regression algorithm is used for model training to construct a fish activity index threshold adjustment model. In actual use, when the system obtains the environmental parameters of the water body in real time, such as the current water temperature is 23 degrees Celsius, the flow rate is 0.3 meters per second, the dissolved oxygen content is 7 milligrams per liter, and the turbidity is 4NTU, these real-time environmental parameters are input into the trained fish activity index threshold adjustment model, and the benchmark fish activity index threshold suitable for the current water conditions is obtained as 7.2. The real-time fish activity index is compared with the adjusted baseline fish activity index threshold. If the difference is less than the difference threshold of 1.0, the health status of the fish school is judged to be normal. If the real-time fish activity index is 6.8, the difference between the real-time fish activity index and the baseline fish activity index threshold of 7.2 is 0.4, which is less than the preset difference threshold of 1.0, so the health status of the fish school is judged to be normal. If the fish activity index is monitored to be 5.5 at a certain moment, the difference with the baseline fish activity index threshold is 1.7, which is greater than the preset difference threshold of 1.0, then the health status of the fish school is judged to be abnormal. The system will immediately sound an alarm, start the drone-mounted camera to take the fish body appearance image of the abnormal fish school, and upload and save the fish body appearance image of the fish school to the fish school appearance image database.

[0051] Step S105, identifying the behavioral characteristics of the school of fish and the external symptoms of the fish based on the appearance images of the fish with abnormal health status, and determining the type of the disease of the school of fish.

[0052] Through the fish school appearance image database, historical fish appearance images with abnormal health status are obtained, and behavioral characteristics and external symptoms are marked. The convolutional neural network is used for model training to build a fish school feature recognition model. Based on the real-time acquired fish appearance images with abnormal health status and fish appearance images of fish schools with a density change rate higher than the preset density threshold, the fish school behavior characteristics and fish external symptoms are identified. The fish school behavior characteristics include but are not limited to abnormal swimming and floating heads, and the fish external symptoms include but are not limited to surface lesions, decolorization, scale abnormalities, ulcers, and parasites. According to the fish school behavior characteristics and fish external symptoms, the fish school disease category is determined. Through the fish school health monitoring equipment, the fish school health status and fish school disease category are sent to the fish school health management center in real time.

[0053] For example, through the fish school appearance image database, 5,000 fish school images with marked behavioral characteristics and external symptoms were obtained, of which 2,000 images showed that the fish school had different degrees of abnormal swimming, such as unstable swimming trajectory and sideways swimming, 1,500 images showed the phenomenon of fish school floating head, and the remaining 1,500 images showed the external symptoms of the fish body, including surface lesions, discoloration, scale abnormalities, ulcers, and parasite attachment. The convolutional neural network was used for model training to build a fish school feature recognition model. By training and verifying the images marked with fish school behavioral characteristics and fish body external symptoms, it was identified whether the fish school had abnormal swimming. The recognition accuracy of abnormal swimming was 93%, and the recognition accuracy of floating head behavior was 90%. For the external symptoms of the fish body, the model had a recognition accuracy of 95% for surface lesions, and the recognition accuracy of other symptoms such as scale abnormalities and discoloration was 92% and 91% respectively. In the actual monitoring process, fish health monitoring equipment collects real-time images of fish with abnormal health status and fish with a density change rate higher than the preset density threshold, and analyzes them through the trained convolutional neural network model. If in a certain monitoring, some fish are identified to have abnormal swimming behavior, and image analysis also finds that 10 of these fish have obvious ulcer characteristics on their body surface. Based on these behavioral characteristics and external symptoms, it is inferred that these fish suffer from typical bacterial ulcer disease, which is usually caused by water quality deterioration or pathogen infection.

[0054] Step S106, formulate fish health intervention measures according to the fish disease category, and send them to the automated water quality control equipment and the automatic drug delivery equipment to manage the fish health, monitor the fish activity index in real time, and judge the effectiveness of the fish health intervention measures.

[0055] Through the dispatching system of the fish health management center, fish health intervention measures are formulated according to the types of fish diseases. The intervention measures include water quality adjustment requirements, required drug types and dosages, and the intervention measures are recorded and sent to the automated water quality control equipment and automatic drug delivery equipment. Oxygen aeration equipment and pH value adjustment equipment are used to adjust water quality parameters, and automatic drug delivery equipment is used to inject drugs into the water body according to the preset delivery rate and cycle according to the disease prevention or treatment requirements in the fish health adjustment work order, and the drug delivery information is transmitted to the fish health management center. The fish activity index is monitored in real time, and the current fish activity index is compared with the benchmark fish activity index threshold to determine whether the fish health intervention measures are effective, and the feedback data is sent to the fish health management center. If the fish health intervention measures are effective and the fish health status returns to normal, the water quality control equipment and automatic drug delivery equipment are shut down through the dispatching system of the fish health management center, and the water quality adjustment and drug delivery are stopped. If the fish health intervention measures are not effective, new fish health intervention measures are generated and the water quality control equipment and automatic drug delivery equipment are readjusted until the fish health status returns to normal.

[0056] For example, if the fish health status in a certain water area is abnormal, it is inferred that these fish may suffer from bacterial canker. Combined with water quality data, it is found that the current dissolved oxygen content in the water is 5 mg / L and the pH value is 6.5, while the optimal water quality conditions for this type of fish are a dissolved oxygen content of not less than 8 mg / L and a pH value between 7.0 and 7.5. Based on the external symptoms of the fish, it is recommended to use antibiotics such as florfenicol for treatment, and the recommended dose is 0.5 grams per cubic meter of water. Based on this, the dispatching system of the fish health management center generates a set of intervention measures, including water quality adjustment and drug treatment, including raising the dissolved oxygen level to at least 8 mg / L, using oxygen aeration equipment to oxygenate the water, adjusting the pH value of the water to 7.0, using pH adjustment equipment to add sodium bicarbonate for neutralization, and using automatic dosing equipment to add florfenicol at a dose of 0.5 grams per cubic meter, and setting the release cycle to once every 12 hours for 3 days. The intervention measures are recorded in the fish health management work order and automatically sent to the corresponding equipment control port. The oxygen enrichment equipment and pH adjustment equipment start to operate according to the set water quality target, and the automatic dosing equipment starts the drug injection process according to the preset dosage and time period. After each dosing, the equipment will transmit the dosing time, dosage and real-time parameters of the water body to the fish health management center. During the intervention process, the activity index of the fish is monitored in real time. If the baseline fish activity index threshold is 7.0, the initial monitored fish activity index is 5.5, which reflects the low activity state of the fish. After 24 hours of the intervention measures, the activity index of the fish increased to 6.8, and the behavior of the fish also tended to be normal, indicating that the intervention measures are initially effective. 48 hours after the intervention measures were implemented, the activity index rose to 7.2, exceeding the baseline activity index threshold of 7.0, and the external symptoms of the fish were significantly alleviated, and the health of the fish returned to normal. The dispatch system of the fish health management center shut down the oxygen aeration equipment, pH adjustment equipment and automatic drug delivery equipment, stopped all water quality adjustment and drug delivery work, and recorded the process and results to the fish health management center to form a complete health intervention report. If the health of the fish fails to improve effectively during the implementation of the intervention, such as the activity index remains at 5.5 or decreases after 24 hours, new intervention measures are generated based on the current environment and health data, such as increasing the dosage of antibiotics, changing the type of drug, or adjusting the water quality target, until the health of the fish returns to normal.

[0057] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the concept of the present application. For example, the above features are replaced with the technical features with similar functions disclosed in the present application (but not limited to) to form a technical solution.

Claims

1. A fish density monitoring method based on big data and acoustic wave signals, characterized in that: The method comprises: According to the sound wave reflection signal of the water area to be detected, a sound wave reflection source type identification model is constructed to determine the type of the sound wave reflection source of the sound wave reflection signal and the position of the sound wave reflection source; According to the acoustic wave reflection signals of the school of fish, the similarity of the frequency spectrum characteristics of the acoustic wave reflection signals of the school of fish is calculated, and the multi-path reflection signals in the same acoustic wave reflection source are identified and eliminated; Based on the echo width, spectrum characteristics, phase difference and phase offset of the fish school acoustic reflection signal, a fish school density prediction model is constructed to predict the fish school density, and the fish school activity index is determined based on the swimming speed, density, swimming speed fluctuation and speed change rate of the fish school in the water area; The fish health status is determined based on the baseline fish activity index threshold and the fish activity index, and the drone-mounted camera is used to obtain fish body appearance images of fish with abnormal health status; Based on the appearance images of fish with abnormal health status, identify the behavioral characteristics and external symptoms of the fish, and determine the type of fish disease; According to the types of fish diseases, fish health intervention measures are formulated and sent to automated water quality control equipment and automatic drug delivery equipment for fish health management, real-time monitoring of fish activity index, and judgment of the effectiveness of fish health intervention measures.

2. The method according to claim 1, wherein: The method of constructing a sound wave reflection source type identification model according to the sound wave reflection signal of the water area to be detected, determining the sound wave reflection source type of the sound wave reflection signal, and determining the position of the sound wave reflection source includes: The acoustic detection equipment emits acoustic waves to the water area to be detected according to the set acoustic emission parameters, and the acoustic reflection signal and the corresponding timestamp information are obtained through the directional array sensor; according to the acoustic reflection signal of the water area to be detected, the acoustic reflection signal is separated by the independent component analysis algorithm to obtain the acoustic reflection signals of different acoustic reflection sources, and saved in the acoustic reflection monitoring database; the time domain acoustic reflection signal is converted into frequency domain data by the short-time Fourier transform algorithm, and its spectrum characteristics are extracted, and the spectrum characteristics include frequency, frequency peak and amplitude; through the acoustic reflection monitoring database, the historical acoustic emission frequency, amplitude, and spectrum characteristics of the acoustic reflection signal are obtained, and the type of acoustic reflection source is marked, and the support vector machine algorithm is used for model training to build an acoustic reflection source type recognition model to determine the acoustic reflection source type of the acoustic reflection signal, including fish school acoustic reflection source and non-fish school acoustic reflection source; the echo time of the acoustic reflection signal of the fish school is obtained, and the position of the fish school relative to the sensor array is calculated by time difference estimation to determine the position of the fish school.

3. The method according to claim 1, wherein: The method of calculating the similarity of the frequency spectrum characteristics of the fish school sound wave reflection signals according to the fish school sound wave reflection signals, identifying and eliminating the multipath reflection signals in the same sound wave reflection source, comprises: The acoustic wave reflection signals of the fish school are obtained through the acoustic wave reflection monitoring database, and the similarity of the spectral characteristics of the acoustic wave reflection signals of the fish school is calculated using the method of calculating the Pearson correlation coefficient. If the similarity is greater than a preset similarity threshold and the echo time difference is less than a preset difference threshold, the acoustic wave reflection signal of the fish school is determined to be a multi-path reflection signal of the same acoustic wave reflection source; the multi-path reflection signals in the same acoustic wave reflection source are eliminated, and the unique acoustic wave reflection signal of the acoustic wave reflection source is retained.

4. The method according to claim 1, wherein: The method comprises building a fish school density prediction model based on the echo width, spectrum characteristics, phase difference and phase offset of the fish school acoustic reflection signal, predicting the fish school density, and determining the fish school activity index according to the swimming speed, density, swimming speed volatility and speed change rate of the fish school in the water area, including: According to the unique sound wave reflection signal of the sound wave reflection source, the phase difference and phase offset of the sound wave transmission signal are determined; based on the echo width, spectral characteristics, phase difference and phase offset of the sound wave reflection signal of the fish school, a recurrent neural network algorithm is used for model training to build a fish school density prediction model to predict the fish school density; the fish school density is continuously monitored. If the fish school density change rate is higher than the preset density threshold, the camera mounted on the drone is used to obtain the fish body appearance image of the fish school and save it to the fish school appearance image database; if the fish school density change rate is lower than the preset density threshold, the echo signals at different time points are compared to judge the position change of the fish school and determine the swimming speed of the fish school; the fish school activity index is determined according to the swimming speed, density, swimming speed volatility and speed change rate of the fish school in the water area.

5. The method according to claim 4, wherein: Determining the fish school activity index according to the swimming speed, density, swimming speed fluctuation and speed change rate of the fish school in the water area includes: According to the swimming speed of the fish school in the water area, the fish school swimming speed fluctuation formula is used Calculate the fluctuation of fish school swimming speed σ V (t), where V(t) is the swimming speed of the fish school at time t, is the average swimming speed in the time period T; according to the swimming speed, density, swimming speed fluctuation and speed change rate of the fish school in the water area, the fish school activity index formula is used Calculate the fish activity index FAI(t), where V(t) is the average swimming speed of the fish, D(t) is the fish density, and σ V (t) is the fluctuation of the swimming speed of the fish school, is the rate of change of the swimming speed of the fish school.

6. The method according to claim 1, wherein: The method of judging the health status of the fish school according to the reference fish school activity index threshold and the fish school activity index, and obtaining the fish body appearance image of the fish school with abnormal health status by using the camera carried by the drone, includes: According to the calculated fish activity index, the water environment parameters corresponding to the detection time of the fish activity index are obtained in real time through the water environment sensor and saved in the fish monitoring database. The water environment parameters include flow rate, dissolved oxygen content, water temperature, and turbidity; according to the water environment parameters obtained in real time, the baseline fish activity index threshold is adjusted; according to the baseline fish activity index threshold and the fish activity index, the health status of the fish is judged, and the fish health status is normal or abnormal; If the difference between the fish school activity index and the benchmark fish school activity index threshold is greater than the preset difference threshold, the health status of the fish school is judged to be abnormal, and the camera carried by the drone is used to obtain the appearance image of the fish body of the fish school with abnormal health status, and save it to the fish school appearance image database.

7. The method according to claim 6, wherein: The step of adjusting the threshold of the benchmark fish activity index according to the water environment parameters obtained in real time includes: Through the fish monitoring database, historical water environment parameters are obtained, and the benchmark fish activity index thresholds corresponding to the historical water environment parameters are marked. The random forest regression algorithm is used for model training to build a fish activity index threshold adjustment model. According to the real-time water environment parameters, the fish activity index threshold adjustment model is used to adjust the benchmark fish activity index threshold. The health status of the fish is judged based on the adjusted benchmark fish activity index threshold and the real-time fish activity index.

8. The method according to claim 1, wherein: The method of identifying the behavioral characteristics and external symptoms of the fish school based on the appearance images of the fish with abnormal health status, and determining the disease category of the fish school, includes: Through the fish appearance image database, historical fish appearance images with abnormal health status are obtained, and behavioral characteristics and external symptoms are marked. The convolutional neural network is used for model training to build a fish feature recognition model; based on the real-time acquired fish appearance images with abnormal health status and the fish appearance images of fish with a density change rate higher than a preset density threshold, the fish behavior characteristics and fish external symptoms are identified. The fish behavior characteristics include but are not limited to abnormal swimming and floating heads, and the fish external symptoms include but are not limited to surface lesions, decolorization, abnormal scales, ulcers, and parasites; the fish disease category is determined based on the fish behavior characteristics and fish external symptoms; and the fish health status and fish disease category are sent to the fish health management center in real time through fish health monitoring equipment.

9. The method according to claim 1, wherein: According to the fish disease category, fish health intervention measures are formulated and sent to the automated water quality control equipment and the automatic drug delivery equipment for fish health management, real-time monitoring of fish activity index, and determination of the effectiveness of fish health intervention measures, including: Through the dispatching system of the fish health management center, fish health intervention measures are formulated according to the types of fish diseases. The intervention measures include water quality adjustment requirements, required drug types and dosages, and the intervention measures are recorded and sent to the automated water quality control equipment and automatic drug delivery equipment; oxygen aeration equipment and pH value adjustment equipment are used to adjust water quality parameters, and automatic drug delivery equipment is used to inject drugs into the water body according to the preset delivery rate and cycle according to the disease prevention or treatment requirements in the fish health adjustment work order, and the drug delivery information is transmitted to the fish health management center; the fish activity index is monitored in real time, and the current fish activity index is compared with the benchmark fish activity index threshold to determine whether the fish health intervention measures are effective, and the feedback data is sent to the fish health management center; if the fish health intervention measures are effective and the fish health status returns to normal, the water quality control equipment and automatic drug delivery equipment are shut down through the dispatching system of the fish health management center, and the water quality adjustment and drug delivery are stopped; if the fish health intervention measures are ineffective, new fish health intervention measures are generated, and the water quality control equipment and automatic drug delivery equipment are readjusted until the fish health status returns to normal.

Images