Self-adaptive feeding regulation and control device and method based on multispectral dynamic water quality detection

Abstract

The invention discloses a self-adaptive feeding regulation and control method and device based on multispectral dynamic water quality detection, and relates to the technical field of aquaculture. The device comprises a feeding regulation and control center, a color correction plate, a dissolved oxygen sensor, a turbidity sensor, an ORP sensor and a PH sensor. The feeding regulation and control center is a main control unit and comprises a radiation correction module, a multispectral camera, a control calculation center and the like. In addition, the invention further provides a method which comprises the steps of radiation correction, color correction, water body recognition, water color judgment and grading, algae content evaluation, water quality evaluation and drug and bait feeding regulation and control. The water quality condition is comprehensively judged by fusing and correcting accurately water body color information and algae information and water quality index information of a multi-element sensor and combining expert experience, and a regulation and control strategy is dynamically provided. On the basis of a water quality evaluation result, a real-time automatic regulation and control strategy can be realized in combination with experience of breeding experts, and the device is matched for adjustment.

Description

Technical Field

[0001] This invention patent relates to the technical field of aquaculture, and in particular to a device and method for water quality detection and adaptive feeding regulation. Background Art

[0002] During the process of aquaculture, controlling the water quality to provide a suitable growth environment for animals is of great significance for improving the survival rate of aquaculture and production efficiency. Among them, the water quality is affected by planktonic microalgae, animal excretion, bait feeding residues, etc., which is directly reflected in the change of water color. In aquaculture environments with different water colors, the contents of dissolved oxygen, turbidity, ORP, etc. are different, which affect the growth and feeding activities of fish and shrimp. For example, the main algae in brown or dark brown water bodies are diatoms, which are suitable for cultivating fattening fish; the dominant algae in green or dark green water bodies are green algae or chlorella, which are suitable for the rapid growth of fish and shrimp; yellow-green water bodies are helpful for the feeding of fish and shrimp.

[0003] At present, the methods and devices for water quality detection are mainly divided into two types. One is to use satellite remote sensing technology to detect water color. These methods are relatively costly and have limited applications. They cannot be applied to situations such as greenhouse aquaculture and are greatly affected by weather. The other is to detect the information such as dissolved oxygen and turbidity of local point areas by detecting the water body (including putting the equipment into the water). The information reflected is relatively limited. At present, the existing technical methods have great limitations and cannot effectively adapt to the dynamic detection and regulation of aquaculture water quality. Summary of the Invention

[0004] In view of the technical problems mentioned in the above background art, an adaptive feeding regulation device and method based on multi-spectral dynamic water quality detection are provided. The present invention is used to realize the water quality detection of aquaculture water bodies, and is used to improve the aquaculture water bodies and provide a suitable aquaculture environment for fish and shrimp.

[0005] The technical means adopted by the present invention are as follows:

[0006] An adaptive feeding regulation device based on multi-spectral dynamic water quality detection, comprising: a water quality monitoring module and a feeding control center; the feeding control center includes: a radiation correction module, a multi-spectral camera module, and a control calculation center; the water quality monitoring module includes: a dissolved oxygen sensor, a turbidity sensor, an ORP sensor, and a PH sensor;

[0007] The radiation correction module includes: 5 groups of spectral illuminometers; among them, one group of central illuminometers faces upward to obtain solar radiation illuminance; the remaining 4 groups of illuminometers are evenly arranged facing four horizontal directions to obtain environmental radiation illuminance.

[0008] Furthermore, the multi-spectral camera module includes multiple groups of camera groups with different spectra; among them, the multi-spectral camera module includes at least one group of RGB camera groups.

[0009] Furthermore, the control and computing center includes: a control unit and a computing unit; wherein, the control unit includes: a rotating shaft, a lifting rod, and a feeding machine; the control unit is used to control the up-and-down rotation of the multispectral camera to adjust the viewing angle, and the control unit monitors the states of different positions.

[0010] The computing unit is an embedded mobile platform, and the computing unit includes: a microprocessor, a 4G module, a Wifi module, and a storage module; the computing unit acquires multi-source sensor data, judges water quality and water color according to the expert system integrated in the system, and sends control instructions to achieve decision-making guidance.

[0011] Furthermore, the rotating shaft and the lifting shaft are used to adjust the camera viewing angle for detecting different positions.

[0012] Furthermore, the feeding machine includes: a feeding box, a motor, a bait outlet, a water quality regulation material outlet, and a water quality regulation material storage unit; wherein, bait and medicine are respectively put out from the bait outlet and the water quality regulation material outlet; the regulation material storage unit is used to store medicine and preparations.

[0013] The present invention also includes a method for detecting aquaculture water quality, comprising the following steps:

[0014] Step 1: Radiometric correction; distinguish and set indoor, outdoor aquaculture environments or artificially set aquaculture environments according to the water body recognition result; and only when the device is used for the first time, perform attenuation coefficient correction in advance.

[0015] Step 2: Color correction; the RGB camera detects the image of the color correction module and performs camera color correction with the built-in standard color calibration model MCC.

[0016] Step 3: Water body recognition; based on the image obtained by the RGB lens in the multispectral camera, according to different application scenarios, use the pre-trained water body-environment deep learning classification model to segment the water body image, establish the corresponding mask, and according to the coordinate change relationship of the multispectral camera group in hardware, segment the multispectral image of the water body part in other band images.

[0017] Step 4: Water color judgment and grading; construct a pre-trained model based on the prior experience of aquaculture experts for qualitative classification.

[0018] Step 5: Algae content assessment; according to the multispectral image after spectral correction, obtain the reflectance value Re λ (x, y) of each pixel point p(x, y) of the water body part image at different band spectra, and then obtain the reflectance Re λ(x, y) distribution histogram; Select the 95% confidence interval of the reflectivity distribution histogram as the effective reflectivity, and calculate the average effective reflectivity as shown, which is used to represent the global reflectivity information:

[0019]

[0020] Among them, Re λ represents the average reflectivity at wavelength λ, and N CI represents the total number of pixels, and Re λ (x, y) represents the reflectivity of the pixel point with coordinates (x, y) at wavelength λ;

[0021] Step 6: According to the relationship between the spectral absorption coefficients of different algae and the algal species, and the remote sensing water phytoplankton inversion model, substitute Re λ into the reflectivity and plankton species model based on the combined inversion of deep learning - transfer learning to obtain the algal species and concentration;

[0022] Step 7: Water quality assessment and drug and feeding regulation.

[0023] Furthermore, in the said Step 1, place the device in the detection environment, place a standard reflectivity whiteboard near the detection object in the camera's view, and simultaneously measure the spectral radiation value of the whiteboard measured by the DN value range of the radiation correction module of the camera to obtain the central attenuation coefficient σ 1 and the surrounding attenuation coefficient σ 2 :

[0024]

[0025] Due to the influence of environmental factors, the water color spectrum is easily interfered. Perform radiation correction under different conditions for different environments:

[0026] A. For the outdoor environment, mainly use the central illuminometer, take the measured DN1 1 value as the whiteboard value, and substitute it into the following correction formula:

[0027]

[0028] Among them, p represents a certain pixel point in the spectral camera, λ represents each band of the multispectral camera, and for the RGB camera group, it is the exposure value. C(p) λ represents the corrected spectral radiance, and I(p) λ represents the spectral radiance before correction; σ 1 represents the attenuation coefficient, v(p) λ represents the spectral radiance of the p pixel point obtained by the camera shooting in the λ band, and v b represents the dark current value or the spectral radiance obtained by the camera shooting the standard blackboard in a closed environment, DN11 Indicates the spectral radiance measured by the central illuminance meter 11;

[0029] B. For the indoor environment, mainly use the surrounding irradiance meter to obtain the average DN 12 value as the whiteboard value, and substitute it into the following correction formula:

[0030]

[0031] Furthermore, in the step 4, the primary color classification labels are constructed as:

[0032] [Brownish water, dark brown water, emerald green water, yellowish green water];

[0033] [Blueish green water, dark green water, grayish green water, blackish brown water, soy sauce water, red turbid water, yellow turbid water, white turbid water, moss water];

[0034] Among them, brownish water, dark brown water, emerald green water and yellowish green water are the water colors of better aquaculture water bodies, and the rest are poor water colors.

[0035] Furthermore, the pre-trained model is established as follows. Correct and extract the water body RGB image to construct a 9-dimensional color feature subset F c :

[0036] F c = [h r h g h b σ r σ g σ b s r s g σ b ;

[0037] Among them, h r , h g , h b respectively represent the histogram feature coefficients of the R, G, and B channels, which are the sum of the 95% confidence intervals in the histogram, that is, the global effective gray level under this channel:

[0038]

[0039] Among them, i represents the channel, N CI represents the number of pixel points within the 95% confidence interval in the water body image, and v(x, y) represents the gray level of the effective pixel points under this channel; σ r , σ g , σ b and s r , s g , s brespectively represent the second - order color moments and third - order color matrices of the R, G, and B channels, respectively representing the color distribution range and symmetry under that channel, and representing the global color distribution situation under that channel;

[0040]

[0041] Taking the color feature sub - vectors as inputs and the color classification labels as outputs, a pre - trained regression model is obtained through training based on the marked dataset of aquaculture experts. Based on the pre - trained model, the corrected images obtained from the measured RGB camera group are used, and the water body part images within the shooting range are segmented. The color feature sub - vectors are extracted and brought into the model for the discrimination and differentiation of water color.

[0042] ,, compared with the prior art, the present invention has the following advantages:

[0043] The present invention combines a modified radiation correction model to standardize RGB images and multispectral images, avoiding differences in the measured water body spectra and colors caused by factors such as lighting in different aquaculture environments, making the model inputs for water quality discrimination more standardized;

[0044] Water color is affected by suspended substances in water such as feed, phytoplankton (these two are the main factors for water color change), zooplankton, etc. There is an association between water color and algae. The specific water quality situation cannot be determined solely by water color, and only qualitative analysis can be carried out. The present invention combines RGB to judge water color and multispectral to evaluate the algal species and algal content (research in the field of remote sensing), and quantitatively evaluates the water quality state from both the appearance and internal causes (from the aspect of water color).

[0045] The device described in the present invention is the most intuitive for water color reaction in water quality assessment, but it is directly related to factors such as dissolved oxygen, pH, etc. The water quality cannot be accurately reflected solely by water color. Combining indicators such as dissolved oxygen, pH, and ORP inside the water body can more comprehensively and quantitatively and accurately reflect the water quality state.

[0046] Based on the water quality assessment results, the present invention can realize real - time automated control strategies in combination with the experience of aquaculture experts and adjust them in cooperation with the device. In addition, the device of the present invention can be used as a multi - data collection base station, which can be used to enrich real - time water quality data, optimize the expert experience system, and improve the control model strategy.

[0047] The device proposed by the present invention meets the requirements of the method of the present invention and can achieve dynamic monitoring and detection. In addition, through the stratified outlets for bait and drugs, the direct influence between drugs and bait is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0049] Figure 1 Schematic diagram of the feeding regulation center of the adaptive feeding regulation device based on multi-spectral dynamic water quality detection of the present invention;

[0050] Figure 2 Schematic diagram of the structure of the radiation correction module of the present invention;

[0051] Among them, 1 is the radiation correction module, 2 is the multi-spectral camera, 3 is the control calculation center, 4 is the rotating shaft, 5 is the lifting shaft, 6 is the bait cover, 7 is the bait box, 8 is the bait outlet, 9 is the regulation material outlet, 10 is the regulation material storage unit, 11 is the central spectral illuminometer, and 12 is the surrounding spectral illuminometer. Detailed implementation manners

[0052] In order to enable those skilled in the art to better understand the solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0053] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order different from those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0054] An adaptive feeding regulation device based on multi-spectral dynamic water quality detection includes: a water quality monitoring module and a feeding regulation center. Among them, the water quality monitoring module at least includes a dissolved oxygen sensor, a turbidity sensor, an ORP sensor, and a PH sensor. Among them, the feeding regulation center is the main control unit, including a radiation correction module, a multi-spectral camera module, and a control calculation center. The structure is as Figure 1 shown, where:

[0055] Radiation correction module 1: As Figure 2 shown, it includes 5 groups of spectral illuminometers. One group of central illuminometers 11 faces upward, and the other four groups of illuminometers 12 face four directions respectively, used to obtain parameters such as ambient light intensity and light flux, adjust the camera exposure time, make the image obtain relatively stable exposure, and be used for spectral radiation correction to reduce the interference with water color judgment;

[0056] Multi-spectral camera 2: It includes an RGB lens, a multi-spectral lens group. The spectral bands include: 440nm, 450nm, 465nm, 630nm, 675nm bands; it cooperates with a color correction plate to obtain the algae content and distribution of the monitored water body; color correction plate, rotating shaft, lifting rod, and feeding machine.

[0057] Control calculation center 3: It includes a control unit and a calculation unit; the control unit is used to control the up and down rotation of the multi-spectral camera to adjust the viewing angle for monitoring different positions, which is composed of a rotating shaft, a lifting rod, and a feeding machine. The calculation unit is an embedded mobile platform, including a microprocessor, a 4G module, a Wifi module, a storage module, etc., used to obtain multi-source sensor data, judge water quality and water color according to the expert system integrated in the system, and send regulation instructions; support remote wireless and wired methods for manual operation and parameter setting.

[0058] Rotating shaft 4: Used to control the left and right rotation of the camera for monitoring different positions;

[0059] Lifting shaft 5: Used to control the up and down movement of the camera, and cooperate with the rotation angle of the camera to detect different positions;

[0060] Feeding machine bait cover 6 and bait box 7: Used to store and feed bait;

[0061] Feeding machine bait outlet 8: Used to spray and feed bait;

[0062] Feeding machine water quality regulation material outlet 9: Used to spray biological materials, organic and inorganic drugs, etc. for regulating water quality and water color;

[0063] Regulation material storage unit 10: Used to store drugs and preparations such as microbial agent EM, organic acid, calcium phosphate, and solid oxygen.

[0064] In addition, the present invention also provides a method for adaptive feeding regulation based on multi-spectral dynamic water quality detection, comprising the following steps:

[0065] (1) Radiometric calibration: According to the water body recognition result, the indoor and outdoor aquaculture environments are distinguished and set, or the aquaculture environment can also be set manually. Only when the device and method are used for the first time, the attenuation coefficient calibration needs to be carried out first. The specific method is to place the device in the detection environment, place a standard reflectance whiteboard near the detection object in the camera's field of view, and measure the spectral radiance value of the whiteboard measured by the radiometric calibration module and the camera at the same time. The central attenuation coefficient σ can be obtained according to the following formulas (1) and (2): 1 and the surrounding attenuation coefficient σ 2 .

[0066]

[0067] Due to the influence of environmental factors, the water color spectrum is easily interfered. Different conditions of radiometric calibration are carried out for different environments:

[0068] A. For outdoor environments such as ponds, since the spectral radiation changes in the near-ground environment are not significant and are mainly affected by the atmospheric environment and sunlight, the central illuminometer 11 is mainly used, and the DN1 1 value measured by it is used as the whiteboard value and substituted into the following calibration formula (3):

[0069]

[0070] where p is a pixel point in the spectral camera, λ is each band of the multi-spectral camera, for the RGB camera group it is the exposure value, C(p) λ is the calibrated spectral radiance, I(p) λ is the spectral radiance before calibration. σ 1 is the attenuation coefficient, v(p) λ is the spectral radiance of the p pixel point obtained by the camera shooting in the λ band, v b is the dark current value or the spectral radiance obtained by the camera shooting a standard blackboard in a closed environment, and DN 11 is the spectral radiance measured by the central illuminometer 11;

[0071] B. For indoor environments such as greenhouse / small shed facilities, the environment is relatively stable, but the lighting conditions may be non-uniform. Therefore, the surrounding irradiance meter 12 is mainly used, and the average DN 12 value measured by it is used as the whiteboard value and substituted into the following calibration formula (4):

[0072]

[0073] (2) Color correction: Place a standard 24-color color correction plate near the device detection object to correct the RGB color. The specific method is that the RGB camera detects the grid boundaries of the color correction plate and performs camera color correction with the built-in standard color calibration model MCC. This step is only carried out before the first use of the device and method, when the usage environment is changed, and during regular maintenance.

[0074] (3) Water body identification: The device provided by the present invention can be applied to pond farming and greenhouse / small greenhouse farming. For the differences in the images obtained under different application scenarios and different camera perspectives, it is necessary to identify and segment the images of the aquaculture water body part; the specific method is that based on the images obtained by the RGB lens in the multispectral camera, combined with the water body-environment deep learning classification model pre-trained according to different application scenarios, it is used to distinguish the water body, aquaculture barrels, earthen pond boundaries, and complex background environments, segment the water body image, and establish the corresponding mask, and segment the multispectral image of the water body part in other band images;

[0075] (4) Water color judgment and grading: The water color judgment and grading need to construct a pre-trained model based on the prior experience of aquaculture experts for qualitative classification. Therefore, the primary color classification labels are constructed as follows:

[0076] [Brown water, tea-brown water, emerald green water, yellow-green water];

[0077] [Blue-green water, dark green water, gray-green water, black-brown water, soy sauce water, red-turbid water, yellow-turbid water, white-turbid water, moss water]

[0078] Among them, brown water, tea-brown water, emerald green water, and yellow-green water are the water colors of better aquaculture water bodies, and the rest are poor water colors.

[0079] The pre-trained model is established as follows: The water body RGB image is corrected and 9-dimensional color feature sub-F is extracted c As formula (5) below

[0080] F c =[h r h g h b σ r σ g σ b s r s g σ b (5)

[0081] h r , h g , h b are the histogram feature coefficients of the R, G, and B channels respectively, which are the sum of the 95% confidence intervals in the histogram and represent the global effective gray levels under these channels. The specific formulas are as follows:

[0082]

[0083] Among them, i represents the channel, and N CI represents the number of pixel points within the 95% confidence interval in the water body image, and v(x, y) represents the gray value of the effective pixel points in this channel.

[0084] σ r σ g σ b and s r s g s b are respectively the second-order color moment and the third-order color matrix of the R, G, and B channels, representing the distribution range and symmetry of the color in this channel respectively, and representing the global color distribution in this channel.

[0085]

[0086] Using the color feature subset as the input and the color classification label as the output, a pre-trained deep learning model is obtained through training based on the labeled dataset of aquaculture experts. In the present invention, the ResNet network is used for training and learning.

[0087] Based on the pre-trained model, according to the corrected images obtained by the measured RGB camera group, and segmenting to obtain the water body partial image within the shooting range, extracting the color feature subset and bringing it into the model for discrimination and distinction of water color. In addition, the input of data accumulates and expands the database and optimizes the robustness of the network.

[0088] (5) Algae content assessment

[0089] According to the multi-spectral image after spectral correction, obtain the reflectance values Re λ (x, y) of each pixel point p(x, y) in the water body partial image under the spectral bands of 440nm, 450nm, 465nm, 630nm, and 675nm, and then obtain the reflectance Re λ (x, y) distribution histogram. Select the 95% confidence interval of the reflectance distribution histogram as the effective reflectance, and calculate the average effective reflectance as shown in formula (9) to represent the global reflectance information:

[0090]

[0091] (6) According to the relationship between the spectral absorption coefficient of different algae and the algal species, and the remote sensing water body phytoplankton inversion model, bring Re λ into the reflectance and plankton species model based on the combined inversion of deep learning - transfer learning to obtain the algal species and content.

[0092] (7) Water quality assessment and drug and feeding regulation

[0093] Based on the above process, three factors of water color, algal species, and algal content are obtained. Combining with dissolved oxygen sensors, pH sensors, turbidity sensors, and ORP sensors, 7 water quality influencing factor indicators are obtained. The multi-indicators are input into the fine regulation system constructed based on the experience of aquaculture experts to determine the water quality status and pest and disease status, provide feeding and medication guidance, and adjust the feeding time and amount according to the drug action time. According to the drug action time, after a certain stage, based on the monitored results of the water quality changes after actual feeding and medication, it is further determined whether to perform secondary water quality adjustment and make adaptive adjustments. Among them, to avoid the influence between bait and drugs such as organic acids, the bait is put from the bait outlet 8, and the water quality regulation materials are put from the outlet 9 (such as Figure 1 ).

[0094] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments. In the above embodiments of the present invention, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways.

[0095] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An adaptive feeding control device based on multi-spectral dynamic water quality detection, characterized in that: include: Water quality monitoring module and feeding control center; The feeding control center includes: a radiation correction module, a multi-spectral camera module and a control calculation center; the water quality monitoring module includes: a dissolved oxygen sensor, a turbidity sensor, an ORP sensor and a PH sensor; The radiation correction module includes: 5 groups of spectral illuminance meters; one group of central illuminance meters faces upwards and is used to obtain solar radiation illuminance; the remaining 4 groups of illuminance meters are evenly arranged in four horizontal directions and are used to obtain environmental radiation illuminance.

2. The adaptive feeding control device based on multi-spectral dynamic water quality detection according to claim 1 is characterized in that: The multi-spectral camera module includes multiple camera groups with different spectrums; the multi-spectral camera module includes at least one RGB camera group.

3. The adaptive feeding control device based on multi-spectral dynamic water quality detection according to claim 1 is characterized in that: The control and computing center includes: a control unit and a computing unit; wherein the control unit includes: a rotating shaft, a lifting rod and a bait throwing machine; the control unit is used to control the multi-spectral camera to rotate up and down to adjust the viewing angle, and realize the monitoring of different locations through the optimal rotation of the rotating shaft; The computing unit is an embedded mobile platform, which includes: a microprocessor, a 4G module, a Wifi module and a storage module; the computing unit obtains multi-source sensor data, and makes water quality and water color judgments based on the expert system integrated in the system, and sends control instructions to achieve decision-making guidance.

4. The adaptive feeding control device based on multi-spectral dynamic water quality detection according to claim 3 is characterized in that: The rotating axis and the lifting axis are used to adjust the camera viewing angle to detect different locations.

5. The adaptive feeding control device based on multi-spectral dynamic water quality detection according to claim 3 is characterized in that: The bait throwing machine includes: a bait throwing box, a motor, a bait outlet, a water quality control material outlet and a water quality control material storage unit; wherein the bait and medicine are respectively thrown in from the bait outlet and the water quality control material outlet; the control material storage unit is used to store medicines and preparations.

6. A method for detecting aquaculture water quality, using the device described in any one of claims 1 to 5, characterized in that: The following steps are involved: Step 1: Radiation correction; Distinguish and set indoor and outdoor aquaculture environments or artificially set aquaculture environments according to the water body identification results; perform attenuation coefficient calibration in advance when and only when the device is used for the first time; Step 2: Color correction: The RGB camera detects the color calibration module image and performs camera color correction with the built-in standard color calibration model MCC; Step 3: Water body identification: Based on the images obtained by the RGB lens in the multispectral camera, according to different application scenarios, the pre-trained water body-environment deep learning classification model is used to segment the water body image and establish the corresponding mask. Based on the coordinate change relationship of the multispectral camera group on the hardware, the multispectral image of the water body part is segmented from the other band images; Step 4: Water color judgment and classification: construct a pre-training model based on the prior experience of aquaculture experts to perform qualitative classification; Step 5: Algae content assessment: According to the spectrally corrected multispectral image, obtain the reflectance value Re of each pixel point p(x, y) in the water part image under different spectral bands λ (x, y), and then the reflectivity Re is obtained λ (x, y) distribution histogram; select the 95% confidence interval of the reflectivity distribution histogram as the effective reflectivity, and calculate the average effective reflectivity as shown to represent the global reflectivity information: Among them, Re λ Represents the average reflectivity at wavelength λ, N CI Represents the total number of pixels, Re λ (x,y) represents the reflectance of the pixel with coordinates (x,y) at wavelength λ; Step 6: According to the relationship between the spectral absorption coefficient of different algae and the algae species and the remote sensing water phytoplankton inversion model, Re λ The reflectivity and plankton species classification model based on deep learning-transfer learning combined inversion was introduced to obtain the algae species and concentrations; Step 7: Water quality assessment and regulation of medication and feeding.

7. A method for detecting aquaculture water quality according to claim 6, characterized in that: In step 1, the device is placed in a detection environment, a standard reflectivity white board is placed near the detection object in the camera viewing angle, and the spectral radiation value of the white board measured by the DN value range camera of the radiation correction module is measured at the same time to obtain the central attenuation coefficient σ1 and the surrounding attenuation coefficient σ2: Due to the influence of environmental factors, the water color spectrum is easily disturbed. Radiation correction is performed under different conditions for different environments: A. For outdoor environments, the central illuminance meter is used as the main measure, and the measured DN 11 The value is used as the whiteboard value and is substituted into the following correction formula: Where p represents a pixel in the spectral camera, λ represents each band of the multispectral camera, and for the RGB camera group it is the exposure value, C(p) λ represents the corrected spectral radiance, I(p) λ represents the spectral radiance before correction; σ1 represents the attenuation coefficient, v(p) λ represents the spectral radiance of pixel p captured by the camera in the λ band, v b Indicates the dark current value or the spectral radiance obtained by the camera shooting a standard blackboard in a closed environment, DN 11 represents the spectral radiance measured by the central illuminance meter 11; B. For indoor environment, the average DN is obtained by using the surrounding radiometer. 12 The value is used to represent the whiteboard value and is substituted into the following correction formula:

8. A method for detecting aquaculture water quality according to claim 6, characterized in that: In step 4, the primary color classification label is constructed as follows: [Brown water, brown water, emerald green water, yellow-green water]; [blue-green water, dark green water, gray-green water, dark brown water, soy sauce water, red turbid water, yellow turbid water, white turbid water, moss water]; Among them, tea-colored water, tea-brown water, emerald green water and yellow-green water are better water colors for aquaculture, and the rest are poor water colors.

9. A method for detecting aquaculture water quality according to claim 6, characterized in that: The pre-training model is established as follows: the RGB image of the water body is corrected and extracted to construct a 9-dimensional color feature sub-F c : F c =[h r h g h b s r s g s b s r s g s b ]; Among them, h r ,h g ,h b Respectively represent the R, G, B channel histogram feature coefficients, which are the sum of the 95% confidence intervals in the histogram, that is, the global effective grayscale under this channel: Where i represents the channel, N CI represents the number of pixels within the 95% confidence interval in the water image, v(x,y) represents the grayscale value of the valid pixel under this channel; σ r , σ g , σ b and r ,s g ,s b Represent the second-order color moment and third-order color matrix of the R, G, and B channels respectively, representing the distribution range and symmetry of the color under the channel, and representing the global color distribution under the channel; The color feature is taken as input and the color classification label is taken as output. A pre-trained regression model is obtained by training according to the labeled data set of aquaculture experts. Based on the pre-trained model, the calibrated image is obtained according to the measured RGB camera group, and the partial image of the water body within the shooting range is obtained by segmentation. The color feature is extracted and brought into the model for water color discrimination.

Images