CN110910067A - Intelligent regulation and control method and system for live fish transportation water quality by combining deep learning and Q-learning - Google Patents
Intelligent regulation and control method and system for live fish transportation water quality by combining deep learning and Q-learning Download PDFInfo
- Publication number
- CN110910067A CN110910067A CN201911166200.3A CN201911166200A CN110910067A CN 110910067 A CN110910067 A CN 110910067A CN 201911166200 A CN201911166200 A CN 201911166200A CN 110910067 A CN110910067 A CN 110910067A
- Authority
- CN
- China
- Prior art keywords
- water quality
- fish
- live fish
- regulation
- water
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 190
- 241000251468 Actinopterygii Species 0.000 title claims abstract description 179
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000013135 deep learning Methods 0.000 title abstract description 8
- 238000013528 artificial neural network Methods 0.000 claims abstract description 34
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims abstract description 27
- 239000011159 matrix material Substances 0.000 claims abstract description 27
- 239000001301 oxygen Substances 0.000 claims abstract description 27
- 229910052760 oxygen Inorganic materials 0.000 claims abstract description 27
- XKMRRTOUMJRJIA-UHFFFAOYSA-N ammonia nh3 Chemical compound N.N XKMRRTOUMJRJIA-UHFFFAOYSA-N 0.000 claims abstract description 25
- 238000011217 control strategy Methods 0.000 claims abstract description 24
- 238000001514 detection method Methods 0.000 claims abstract description 18
- 230000001276 controlling effect Effects 0.000 claims abstract description 10
- 230000001105 regulatory effect Effects 0.000 claims abstract description 8
- 230000008569 process Effects 0.000 claims description 16
- 230000009471 action Effects 0.000 claims description 12
- 238000013507 mapping Methods 0.000 claims description 11
- 230000007774 longterm Effects 0.000 claims description 5
- 238000011176 pooling Methods 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000003908 quality control method Methods 0.000 claims description 3
- 230000009182 swimming Effects 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 claims description 2
- 230000003213 activating effect Effects 0.000 claims 1
- 230000006870 function Effects 0.000 description 15
- 238000011156 evaluation Methods 0.000 description 5
- QGZKDVFQNNGYKY-UHFFFAOYSA-N Ammonia Chemical compound N QGZKDVFQNNGYKY-UHFFFAOYSA-N 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 238000013441 quality evaluation Methods 0.000 description 4
- 238000009360 aquaculture Methods 0.000 description 3
- 244000144974 aquaculture Species 0.000 description 3
- 238000013145 classification model Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000004083 survival effect Effects 0.000 description 3
- 208000027418 Wounds and injury Diseases 0.000 description 2
- 238000010521 absorption reaction Methods 0.000 description 2
- 229910021529 ammonia Inorganic materials 0.000 description 2
- 230000006378 damage Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 208000014674 injury Diseases 0.000 description 2
- 238000003062 neural network model Methods 0.000 description 2
- 239000002352 surface water Substances 0.000 description 2
- 206010003497 Asphyxia Diseases 0.000 description 1
- 206010010071 Coma Diseases 0.000 description 1
- 206010021143 Hypoxia Diseases 0.000 description 1
- 241000404975 Synchiropus splendidus Species 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 210000001339 epidermal cell Anatomy 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 208000015181 infectious disease Diseases 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 238000009372 pisciculture Methods 0.000 description 1
- 239000012629 purifying agent Substances 0.000 description 1
- 230000000241 respiratory effect Effects 0.000 description 1
- 230000029058 respiratory gaseous exchange Effects 0.000 description 1
- 230000035939 shock Effects 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000003911 water pollution Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
- G06Q10/083—Shipping
- G06Q10/0832—Special goods or special handling procedures, e.g. handling of hazardous or fragile goods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Development Economics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Educational Administration (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Health & Medical Sciences (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Multimedia (AREA)
- Game Theory and Decision Science (AREA)
- Farming Of Fish And Shellfish (AREA)
Abstract
The invention discloses a method and a system for intelligently regulating and controlling the water quality of live fish transportation by combining deep learning and Q-learning. The method comprises the following steps: fitting the information of ammonia nitrogen, dissolved oxygen and pH of the real-time captured water body through a RBF neural network, outputting the water quality grade, and establishing a water quality grading model; positioning and identifying the live fish through a deep neural network based on a Faster-rcnn framework, and establishing a Faster-rcnn live fish detection model; constructing a strategy Q matrix according to the timely reward value and the long-distance return value corresponding to the live fish transport water quality to obtain an optimal regulation and control strategy of the live fish transport water quality, and establishing a Q-learning model; and finally, deploying the Q-learning model, the water quality grading model and the Faster-rcnn-based live fish detection model to a server to dynamically regulate and control the water quality. The invention creatively integrates the deep learning method and the Q-learning method into the intelligent regulation and control of the water quality of live fish transportation, provides a set of precise live fish transportation water quality regulation and control scheme, constructs a live fish transportation water quality regulation and control system, and provides powerful guarantee for the safety of live fish transportation.
Description
Technical Field
The invention relates to intelligent control in aquaculture, in particular to an intelligent regulation and control method and system for water quality in live fish transportation.
Background
The transportation of live fish is an indispensable link in the development process of the fish farming industry, and the introduction of new species, the collection of parent fish and the supply of aquatic products in the fish market can not leave the transportation of live fish. Due to the limitation of transportation conditions, the transportation link of live fishes is easily affected by factors such as water quality deterioration, injury infection, stress and the like, so that the survival rate of the live fishes is reduced, the injury rate is increased (the loss rate of long-distance transportation is more than 10%), the risk is brought to the quality safety of the fishes, and the development of aquaculture industry is restricted. Therefore, the central problem in the transportation of live fish, an important production process, is how to improve the survival rate of the fish.
In a modern live fish transportation system, water quality evaluation is an important link for preventing and controlling water pollution. The water quality problem in the live fish transportation process has an inseparable relation with ammonia nitrogen (TAN) concentration, Dissolved Oxygen (DO) concentration and pH value (pH value). In the transportation process, when ammonia discharged by fishes in water exceeds 12mg/L, the phenomena of difficult breathing, coma, even death and the like can be caused after nonionic ammonia enters the fish bodies; when the concentration of dissolved oxygen in the water body is lower than the suffocation point of the fishes, the fishes die in large numbers due to oxygen deficiency; the pH value of the water body suitable for the fishes is 6.5-8.5, and if the pH value is in an abnormal range for a long time, epidermal cells of respiratory organs of the fishes are damaged, so that the oxygen absorption capacity of the fishes is reduced. The traditional water quality evaluation method usually adopts evaluation methods such as single-factor or linear function models, the water transportation environment is actually a complex system formed by interlacing various factors, and the traditional method is too single and cannot comprehensively and objectively carry out comprehensive evaluation. Therefore, the reasonable evaluation of the water quality in the process of live fish transportation to ensure the safety of the transportation water environment is an urgent problem to be solved.
At present, the research on the intelligent evaluation or regulation of the water quality in live fish transportation is less. A water quality monitoring system of a live fish transport case is only researched by Hongyanqian and the like, the temperature, DO and pH in the case are detected through sensors, then the detected values are sent to a PLC to be compared with set values, and finally corresponding control signals are obtained to drive an actuating mechanism to operate so as to control the water quality. Although the system improves the survival rate of the fish to a certain extent, the system still has the defects of complex operation, low efficiency, time and labor consumption and the like. Therefore, an efficient intelligent water quality control system is urgently needed to accurately control the water quality during live fish transportation and provide powerful safety guarantee for live fish transportation.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the problem of low efficiency of the prior art, the invention provides an intelligent regulation and control system for the water quality of live fish transportation by combining deep learning and Q-learning, wherein a deep learning method and a Q-learning method are innovatively combined on the intelligent regulation and control of the water quality of live fish transportation, and the system aims to solve the defects of complex operation, time consumption, labor consumption and the like of the traditional water quality monitoring system.
The technical scheme is as follows: according to a first aspect of the invention, the invention provides an intelligent regulation and control method for the water quality in live fish transportation, which comprises the following steps:
capturing ammonia nitrogen, dissolved oxygen and pH information of a water body in the transport box in real time, fitting through a RBF neural network, and outputting a water quality grade;
capturing a fish body image in the transport case, and identifying the state of the fish body through a deep neural network based on a Faster-rcnn framework;
acquiring the vibration frequency of a transport vehicle, and obtaining the optimal regulation and control strategy of the current live fish transport water quality according to a Q matrix of a pre-constructed Q-learning model by combining water quality grade information and fish body state information;
and correspondingly regulating and controlling the water quality of the transport box according to the output of the Q matrix.
According to a second aspect of the invention, an intelligent regulation and control system for water quality in live fish transportation is provided, which comprises:
the water quality detection module is used for capturing information of ammonia nitrogen, dissolved oxygen and pH of the water body in the transport box in real time, fitting through a RBF neural network and outputting a water quality grade;
the fish body state detection module is used for capturing a fish body image in the transport case and identifying the fish body state through a deep neural network based on a Faster-rcnn framework;
the water quality regulation and control strategy selection module is used for acquiring the vibration frequency of the transport vehicle, and obtaining the optimal regulation and control strategy of the current live fish transport water quality according to a Q matrix of a pre-constructed Q-learning model by combining water quality grade information and fish body state information;
and the execution module is used for correspondingly regulating and controlling the water quality of the transport box according to the output of the Q matrix.
Has the advantages that:
1. the invention creatively integrates the deep learning method and the Q-learning method into the intelligent regulation and control of the water quality of live fish transportation, provides a set of precise live fish transportation water quality regulation and control scheme, constructs a live fish transportation water quality regulation and control system, can intelligently regulate and control the water quality, and provides powerful guarantee for the safety of live fish transportation.
2. The transportation water quality evaluation is carried out by using the neural network, so that the artificial subjective assumption is avoided, the judgment accuracy is improved, and powerful technical support can be provided for the transportation of the live fish.
3. The multi-mode information fusion is realized by fusing the image characteristics and the parameter information of the water quality, and finally the decision judgment is carried out by Q-learning, so that the method has strong adaptability and stronger control capability compared with a model only containing one kind of information.
4. The components of the system can be simply installed on the existing equipment, the operation is simple, the cost is low, and the system has high practicability.
Drawings
FIG. 1 is a flow chart of the intelligent regulation and control method for water quality in live fish transportation according to the invention;
FIG. 2 is a schematic diagram of a system for monitoring and controlling water quality in real time in accordance with the present invention;
FIG. 3 is a view showing the structure of an RBF neural network used for water quality evaluation;
FIG. 4 is a schematic diagram of a Faster-rcnn based convolutional neural network for locating and identifying live fish.
Detailed Description
The technical scheme of the invention is further explained by combining the attached drawings.
Referring to fig. 1 and 2, the method captures ammonia nitrogen, dissolved oxygen and pH information of a water body in a live fish transport box in real time through ammonia nitrogen, dissolved oxygen and pH sensors, performs fitting, outputs water quality grades, and establishes a water quality grading model through an RBF neural network; installing a camera above the live fish transport case, positioning and identifying the live fish through a deep neural network based on a Faster-rcnn architecture, and establishing a live fish detection model; constructing a strategy Q matrix according to the timely reward value and the long-distance return value corresponding to the live fish transport water quality to obtain an optimal regulation and control strategy of the live fish transport water quality, and establishing a Q-learning model; in practical application, the Q-learning model, the water quality grading model and the live fish detection model are deployed on a server to dynamically regulate and control water quality. The method integrates the image characteristics and the parameter information of the water quality, carries out decision judgment through Q-learning, and has stronger adaptability and control capability compared with a single information model.
The establishment process of the water quality classification model comprises the following steps: live fish is taken and placed on a vibration table for simulated transportation, clean and pollution-free transportation water is put in, and the water quality meets the standard requirements of fishery water. And capturing ammonia nitrogen, dissolved oxygen and pH information of the water body in real time by using ammonia nitrogen, dissolved oxygen and pH sensors, fitting the sampled information through a RBF neural network, and outputting the water quality grade. The RBF neural network is trained as follows: the method comprises the steps that a concentration value of ammonia nitrogen (TAN), a concentration value of Dissolved Oxygen (DO) and a pH value are obtained after ammonia nitrogen, dissolved oxygen and pH sensors are used for real-time capture, a neural network takes the ammonia nitrogen, the dissolved oxygen and the pH value as input, then an activation function is used for mapping the neural network to a high-dimensional space, and a real water quality grade is used as a real value of the network, so that the network is continuously updated in an iterative mode until the maximum iteration times or the mean square error meets the preset condition. The water quality grade in the invention comprises five grades. Firstly, dividing the water quality into 5 classes according to the water quality grade division standard of 'standard limit value of surface water environment quality standard basic project', using the 5 classes as a label for network learning, and generating different water quality data within the interval of the class to train an RBF neural network. The trained RBF neural network model is utilized, and then water indexes acquired in real time are input, so that the grades of different water qualities can be identified, the neural network evaluation water quality grade has the advantages that the water quality grade can be divided in a nonlinear mode to the maximum extent, water quality characteristics are mapped to a high-dimensional space to be divided, and the higher latitude is, the more abstract the water quality is, the easier the classification is, and the figure 3 is shown.
The process of establishing the live fish detection model comprises the following steps: the camera is installed in the water tank, the target fish body is positioned and identified in real time through a deep neural network based on a Faster-rcnn framework, and the fish body is judged to be in a normal floating state or face upward to the water surface according to the figure 4, and the method specifically comprises the following steps:
(1) collecting images of fish bodies, and calibrating the coordinate positions of live fish swimming normally and fish bodies turning upwards on the water surface;
(2) and sending the calibrated image into a Faster-rcnn network model. The fast-rcnn firstly extracts image information through a self convolution layer, namely the characteristics of images of fishes in a transport case are extracted, then the images are sent to the RPN, the output of the RPN is regressed and classified through a pooling layer and a connecting layer, which kind of fishes and the coordinate positions of the fishes are calculated, namely the RPN simultaneously executes two functions, firstly, whether pixel blocks in a scene are fishes is identified through the classification function, further, whether the fishes are dead fishes or live fishes is identified, and secondly, the coordinate positions of the fishes are given through the regression function. Where the error is regressed back to its coordinate position by gradient descent, with the classification function being performed by the softmax function. The softmax function is described as follows:
where θ 1, θ 2 are parameters of the Faster-rcnn model. The classification function is performed by the estimated parameters. P is the probability of dead or live fish, x(i)Is to acquire the ith original image sample, y(i)Is the category corresponding to the ith input sample, there are three possibilities: dead fish, live fish, or both,the sum of the probability distributions is made 1, which is normalized. In the invention, the fish body turning upwards is considered as dead fish, the fish body swimming normally is considered as live fish, and when the fish body is identified as dead fish, the storage unit of the system is added with 1, and the number of the dead fish is recorded.
The Q-learning model establishing process comprises the following steps: the vibration frequency is first acquired. In practical applications, the vibration frequency is obtained by mounting a vibration sensor on the transport vehicle. The different vibration frequencies are used for simulating the vibration condition of the transport vehicle when transporting fish, and the vibration frequencies generated by different road conditions are different. The vibration of the transport vehicle has a great influence on the water quality and the life activities of the fish. Generally, the higher the vibration frequency, the stronger the stress on the fish and the poorer the water quality. Researches show that with the enhancement of vibration frequency (five groups of 0Hz, 50Hz, 100Hz, 150Hz and 200 Hz), the influence on the fish is more obvious, the ammonia nitrogen and the pH value of the water body are also increased, and the water quality is obviously reduced, which can be seen in the text of the basic research on mandarin fish living body transportation technology of 2015 Cheng Kun university of oceans. The frequency of the vibration is simulated in the examples by adjusting the vibration frequency of the rocking platform such that the vibration frequency of the rocking platform is from 0Hz to 300Hz, and a level is output every 10Hz for a total of 30 levels of vibration frequency randomly. According to the invention, through simulating the frequency of the transport vehicle, the model can provide different control strategies according to different vibration frequencies, so that an optimal water quality environment is provided for the fish, and because the frequency is continuously changed, all possible frequencies under actual conditions can be generated, namely all possible water quality adjustment strategies are formed, and further the method is applied to actual scenes.
Forming a triple group of the generated 5 kinds of water quality state information, 2 kinds of live fish motion state information and 30 kinds of frequency state information as a state to form a transport environment state matrix, wherein 300 kinds of state matrices are in total; and constructing a strategy matrix Q matrix according to the timely reward value and the long-term return value corresponding to the water quality of the live fish transportation, and obtaining the optimal regulation and control strategy of the water quality of the live fish transportation through the Q matrix.
The Q function is defined as follows: q (s, a) ═ r (s, a) + γ max { r (s ', a') }, wherein Q (s, a) represents a mapping function of the water quality of live fish transportation from a triple s formed by the water quality grade, the fish body state and the vibration frequency to the action a, r (s, a) represents timely reward after the action mapping is completed, γ max { r (s ', a') } represents long-term return, max { r (s ', a') represents a maximum timely reward value which can be generated at the next moment, γ represents attenuation of a future reward value, s 'represents the next possible triple state, and a' represents the next possible water quality regulation and control action. And after the Q function is calculated once, updating the strategy matrix for regulating and controlling the transport water quality once, and stopping updating when the strategy matrix is converged. Firstly, a score (namely a Q value) is obtained according to different frequencies, water quality grades and states (numbers) of fishes, the water quality grades are continuously changed along with the continuous change of the vibration frequencies, the states of the fishes are continuously changed, the score is also continuously changed, a strategy is obtained, the changes tend to be stable after continuous iteration, and a strategy matrix is further obtained. In practical application, given any one of the current water quality, the fish state and the vibration frequency, the other value can be obtained through the strategy matrix, the aquarium is adjusted to the proper vibration frequency through the vibration equipment and the shock absorption equipment, or the excellent water quality grade is achieved through the water quality adjusting device, and the fish is guaranteed to be kept in the best state in the transportation process.
The timely reward related to the water quality control system for live fish transportation is defined as r (s, a) ═ 1-lambda) (η -mu), wherein lambda represents the number of dead fish, η represents the water quality grade and is mu vibration frequency, the physical quantities are all performed after de-dimensionalization, and the formula of de-dimensionalization is many and can be selected from one and is not detailed.
According to another embodiment of the invention, an intelligent regulation and control system for water quality in live fish transportation is provided, which comprises:
the water quality detection module is used for capturing information of ammonia nitrogen, dissolved oxygen and pH of the water body in the transport box in real time, fitting through a RBF neural network and outputting a water quality grade;
the fish body state detection module is used for capturing a fish body image in the transport case and identifying the fish body state through a deep neural network based on a Faster-rcnn framework;
the water quality regulation and control strategy selection module is used for acquiring the vibration frequency of the transport vehicle, and obtaining the optimal regulation and control strategy of the current live fish transport water quality according to a Q matrix of a pre-constructed Q-learning model by combining water quality grade information and fish body state information;
and the execution module is used for correspondingly regulating and controlling the water quality of the transport box according to the output of the Q matrix.
Specifically, the water quality detection module comprises a sensor unit and an RBF neural network computing unit, and further the sensor unit comprises an ammonia nitrogen sensor, a dissolved oxygen sensor and a pH sensor which are respectively used for capturing ammonia nitrogen, dissolved oxygen and pH information of a water body in the live fish transport box in real time; then sending the water quality to an RBF neural network computing unit, using ammonia nitrogen, dissolved oxygen and pH value as input, mapping the water quality to a high-dimensional space by using an activation function, and using real water quality as the real value of the network, so that the network is continuously updated in an iterative manner, and further the grades of different water qualities can be identified. The water quality grade in the invention comprises five grades. Firstly, dividing the water quality into 5 classes according to a water quality grade division standard of 'standard limit value of surface water environment quality standard basic project', using the 5 classes as a label for network learning, and generating different water quality data within grade intervals to train an RBF neural network. And (3) by utilizing the trained RBF neural network model and inputting the water body indexes acquired in real time, the grades of different water qualities can be identified.
The fish body state detection module comprises an image capturing unit and an image calculating unit, further, the image capturing unit captures images of fish bodies in the transport box by using a camera and sends the images to the image calculating unit, and the image calculating unit identifies the fish body states based on a deep neural network of a Faster-rcnn framework to obtain which type of fish (dead fish or live fish) and the coordinate position in the images. The process is as follows: (1) calibrating the coordinate positions of the fish body which swims normally and the fish body which turns upwards on the water surface for the image of the fish body; (2) inputting the calibrated image into a fast-rcnn network model, extracting the characteristics of the image through a convolution layer, then sending the image into an RPN network, performing regression and classification on the output of the RPN network through a pooling layer and a connecting layer, calculating which type of fish and the coordinate position thereof, wherein the coordinate position thereof is regressed in a gradient descent and error back propagation mode, and performing a classification function through a softmax function.
The water quality regulation and control strategy selection module comprises a frequency sensing unit and a regulation and control strategy decision unit, further, the frequency sensing unit senses the vibration frequency of the transport case through a vibration sensor and sends the vibration frequency to the regulation and control strategy decision unit, the regulation and control strategy decision unit also obtains the water quality grade and the fish body state at the same time, an environment state matrix is constructed by combining the vibration frequency, and the optimal regulation and control strategy of the current live fish transport water quality is obtained through a pre-obtained Q matrix.
The Q matrix is obtained according to the following Q-learning model:
Q(s,a)=r(s,a)+γmax{r(s’,a’)}
q (s, a) represents a mapping function from a triple s formed by water quality grade, fish body state and vibration frequency to action a of the transport water quality, r (s, a) represents timely reward after action mapping is completed, gamma max { r (s ', a') } represents long-term return, max { r (s ', a') represents a maximum timely reward value which can be generated at the next moment, gamma represents attenuation of a future reward value, s 'represents the next possible triple state, and a' represents the next possible water quality regulation action.
The timely reward is defined as r (s, a) ═ 1- λ (η - μ), where λ represents the number of dead fish, η represents the water quality grade, and is μ vibration frequency, all of the above physical quantities are subjected to de-dimensionalization, and the formula for de-dimensionalization is many and one of them can be selected, and will not be described in detail.
The execution module utilizes the water quality adjusting device to correspondingly regulate and control the water quality of the transport box according to the output of the Q matrix, for example, a water quality purifying agent, a water changing device, an oxygen increasing device and the like can be used for adjusting the Ph, ammonia nitrogen and dissolved oxygen concentration of water.
In the practical application process, the RBF neural network computing unit, the image computing unit and the regulation and control strategy decision unit can be realized by using separate computing devices, and can also be integrated in one device to realize operation. Referring to fig. 2, in the embodiment, a trained water quality classification model, a live fish detection model and a Q-learning model are deployed on a server, wherein the water quality classification model can automatically evaluate water quality and output a water quality grade; the live fish detection model can detect the state of live fish in the transportation process in real time; the Q-learning model can automatically analyze and obtain the optimal regulation and control strategy of the water quality in the live fish transportation process, and then dynamically regulate and control the water quality. The method is very simple and convenient to operate, only simple installation is needed on the existing equipment, the cost can be greatly saved, the phenomenon that the live fish die in a large number can be effectively avoided, powerful guarantee is provided for the transportation of the live fish, and the economic benefit is greatly improved.
The invention provides a live fish transportation water quality intelligent regulation and control system combining deep learning and Q-learning, which can monitor main water quality parameters and the state of live fish in real time in the transportation process, and when the water quality level is higher or lower than a normal range or the live fish turns upwards, the system can calculate the optimal regulation and control strategy of the water quality in the live fish transportation process according to the obtained parameters so as to automatically regulate the water quality. The system provides a set of scheme for accurately regulating and controlling the quality of the live fish transportation water, realizes the nondestructive detection of aquaculture industry, avoids the phenomenon that live fish dies in a large range, effectively ensures the healthy development of the live fish transportation industry, and can greatly improve the economic benefits of farmers.
Claims (10)
1. An intelligent regulation and control method for water quality in live fish transportation is characterized by comprising the following steps:
s1, capturing ammonia nitrogen, dissolved oxygen and pH information of a water body in the transport box in real time, and giving a corresponding water quality grade by using a pre-constructed water quality grading model;
s2, capturing the fish body image in the transport case, and identifying the fish body state through an image identification processing technology;
s3, obtaining the vibration frequency of the transport vehicle, and obtaining the optimal regulation and control strategy of the current live fish transport water quality according to a Q matrix of a pre-constructed Q-learning model by combining the water quality grade information and the fish body state information;
and S4, correspondingly regulating and controlling the water quality of the transport box according to the output of the Q matrix.
2. The intelligent regulation and control method for water quality in live fish transportation according to claim 1, wherein the step S1 is to output the water quality grade through a pre-trained RBF neural network, and the training process comprises: ammonia nitrogen, dissolved oxygen and pH information of a water body in the transport box are captured in real time through an ammonia nitrogen sensor, a dissolved oxygen sensor and a pH sensor respectively; inputting ammonia nitrogen, dissolved oxygen and pH value into RBF neural network, mapping the RBF neural network to high-dimensional space by using activating function, and performing iterative update by using real water quality grade as real value of the network until the maximum number of iterations is satisfied or the mean square error satisfies the preset condition.
3. The intelligent regulation and control method for water quality in live fish transportation according to claim 1, wherein the step S2 is to identify the fish body state based on a deep neural network of fast-rcnn architecture, and comprises:
(1) collecting images of fish bodies, and calibrating the coordinate positions of the fish bodies in normal swimming and the fish bodies turned upwards on the water surface;
(2) inputting the calibrated image into a fast-rcnn network model, extracting the characteristics of the image through a convolution layer, then sending the image into an RPN network, performing regression and classification on the output of the RPN network through a pooling layer and a connecting layer, calculating the coordinate position of a live fish or a dead fish and the coordinate position of the live fish or the dead fish, wherein the coordinate position of the live fish or the dead fish is regressed in a gradient descent and error back propagation mode, and performing a classification function through a softmax function.
4. The intelligent regulation and control method for water quality in live fish transportation according to claim 1, wherein the pre-constructed Q-learning model is as follows:
Q(s,a)=r(s,a)+γmax{r(s’,a’)}
q (s, a) represents a mapping function from a triple s formed by water quality grade, fish body state and vibration frequency to action a of the transport water quality, r (s, a) represents timely reward after action mapping is completed, gamma max { r (s ', a') } represents long-term return, max { r (s ', a') represents a maximum timely reward value which can be generated at the next moment, gamma represents attenuation of a future reward value, s 'represents the next possible triple state, and a' represents the next possible water quality regulation action.
5. The intelligent control method for the water quality in live fish transportation according to claim 4, wherein the timely reward is defined as r (s, a) ═ 1- λ (η - μ), wherein λ represents the number of dead fish, η represents the water quality grade, and μ is the vibration frequency.
6. The utility model provides a quality of water intelligent regulation and control system is transported to live fish which characterized in that includes:
the water quality detection module is used for capturing information of ammonia nitrogen, dissolved oxygen and pH of the water body in the transport box in real time, fitting through a RBF neural network and outputting a water quality grade;
the fish body state detection module is used for capturing a fish body image in the transport case and identifying the fish body state through a deep neural network based on a Faster-rcnn framework;
the water quality regulation and control strategy selection module is used for acquiring the vibration frequency of the transport vehicle, and obtaining the optimal regulation and control strategy of the current live fish transport water quality according to a Q matrix of a pre-constructed Q-learning model by combining water quality grade information and fish body state information;
and the execution module is used for correspondingly regulating and controlling the water quality of the transport box according to the output of the Q matrix.
7. The intelligent regulation and control system for water quality in live fish transportation according to claim 6, wherein the water quality detection module comprises a sensor unit and an RBF neural network computing unit, wherein the sensor unit comprises an ammonia nitrogen sensor, a dissolved oxygen sensor and a pH sensor, and is used for capturing ammonia nitrogen, dissolved oxygen and pH information of a water body in the transportation box in real time respectively; and then sending the water quality to an RBF neural network computing unit, wherein the RBF neural network computing unit takes ammonia nitrogen, dissolved oxygen and pH value as input, maps the input to a high-dimensional space by using an activation function, continuously iteratively updates by taking the real water quality grade as the real value of the network, and identifies the grades of different water qualities.
8. The intelligent regulation and control system for water quality in live fish transportation according to claim 6, wherein the fish body state detection module comprises an image capturing unit and an image calculating unit, the image capturing unit captures images of fish bodies in the transportation box and sends the images to the image calculating unit, and the image calculating unit identifies the fish body state based on a deep neural network of a Faster-rcnn architecture, and the process is as follows: (1) calibrating the coordinate positions of the fish body which swims normally and the fish body which turns upwards on the water surface for the image of the fish body; (2) inputting the calibrated image into a fast-rcnn network model, extracting the characteristics of the image through a convolution layer, then sending the image into an RPN network, performing regression and classification on the output of the RPN network through a pooling layer and a connecting layer, calculating which type of fish and the coordinate position thereof, wherein the coordinate position thereof is regressed in a gradient descent and error back propagation mode, and performing a classification function through a softmax function.
9. The system as claimed in claim 6, wherein the water quality control strategy selection module comprises a frequency sensing unit and a control strategy decision unit, the frequency sensing unit senses the vibration frequency of the transport box through a vibration sensor and sends the vibration frequency to the control strategy decision unit, the control strategy decision unit obtains the water quality grade and the fish body state, an environment state matrix is constructed by combining the vibration frequency, and the optimal control strategy of the current water quality of live fish transport is obtained through a pre-obtained Q matrix.
10. The intelligent regulation and control system of quality of water of live fish transportation of claim 9, wherein the Q matrix is obtained according to the following Q-learning model:
Q(s,a)=r(s,a)+γmax{r(s’,a’)}
q (s, a) represents a mapping function from a triple s formed by water quality grade, fish body state and vibration frequency to action a of the transport water quality, r (s, a) represents timely reward after action mapping is completed, gamma max { r (s ', a') } represents long-term return, max { r (s ', a') represents a maximum timely reward value which can be generated at the next moment, gamma represents attenuation of a future reward value, s 'represents the next possible triple state, and a' represents the next possible water quality regulation action.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911166200.3A CN110910067A (en) | 2019-11-25 | 2019-11-25 | Intelligent regulation and control method and system for live fish transportation water quality by combining deep learning and Q-learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911166200.3A CN110910067A (en) | 2019-11-25 | 2019-11-25 | Intelligent regulation and control method and system for live fish transportation water quality by combining deep learning and Q-learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110910067A true CN110910067A (en) | 2020-03-24 |
Family
ID=69819300
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911166200.3A Pending CN110910067A (en) | 2019-11-25 | 2019-11-25 | Intelligent regulation and control method and system for live fish transportation water quality by combining deep learning and Q-learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110910067A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111738139A (en) * | 2020-06-19 | 2020-10-02 | 中国水产科学研究院渔业机械仪器研究所 | Cultured fish monitoring method and system based on image recognition |
CN112772485A (en) * | 2021-01-15 | 2021-05-11 | 广西壮族自治区水产科学研究院 | Ecological prawn breeding method |
CN113505649A (en) * | 2021-06-10 | 2021-10-15 | 广州杰赛科技股份有限公司 | Tap water chlorination control method and device |
CN114136906A (en) * | 2021-12-01 | 2022-03-04 | 浙江省海洋水产养殖研究所 | Intelligent fishery regulation and control method and system based on hyperspectral fish meat quality detection |
CN114620819A (en) * | 2022-03-01 | 2022-06-14 | 红云红河烟草(集团)有限责任公司 | Method for adjusting pH value of circulating water for spraying and washing cigarette peculiar smell gas |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160117587A1 (en) * | 2014-10-27 | 2016-04-28 | Zhicheng Yan | Hierarchical deep convolutional neural network for image classification |
CN106296437A (en) * | 2016-09-05 | 2017-01-04 | 华中农业大学 | A kind of transportation of live fish information system, method for building up and application |
CN107156020A (en) * | 2017-06-21 | 2017-09-15 | 重庆大学 | A kind of Intelligent fish tank water quality adjustment method based on intensified learning |
CN107743913A (en) * | 2017-10-13 | 2018-03-02 | 南京师范大学 | A kind of new Pelteobagrus fulvidraco transportation resources based on intelligent control |
CN109543679A (en) * | 2018-11-16 | 2019-03-29 | 南京师范大学 | A kind of dead fish recognition methods and early warning system based on depth convolutional neural networks |
CN110476839A (en) * | 2019-07-24 | 2019-11-22 | 中国农业大学 | A kind of optimization regulating method and system based on fish growth |
-
2019
- 2019-11-25 CN CN201911166200.3A patent/CN110910067A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160117587A1 (en) * | 2014-10-27 | 2016-04-28 | Zhicheng Yan | Hierarchical deep convolutional neural network for image classification |
CN106296437A (en) * | 2016-09-05 | 2017-01-04 | 华中农业大学 | A kind of transportation of live fish information system, method for building up and application |
CN107156020A (en) * | 2017-06-21 | 2017-09-15 | 重庆大学 | A kind of Intelligent fish tank water quality adjustment method based on intensified learning |
CN107743913A (en) * | 2017-10-13 | 2018-03-02 | 南京师范大学 | A kind of new Pelteobagrus fulvidraco transportation resources based on intelligent control |
CN109543679A (en) * | 2018-11-16 | 2019-03-29 | 南京师范大学 | A kind of dead fish recognition methods and early warning system based on depth convolutional neural networks |
CN110476839A (en) * | 2019-07-24 | 2019-11-22 | 中国农业大学 | A kind of optimization regulating method and system based on fish growth |
Non-Patent Citations (1)
Title |
---|
谢万里: "人工神经网络在活鱼运输中水质评价的作用", 《江苏农业科学》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111738139A (en) * | 2020-06-19 | 2020-10-02 | 中国水产科学研究院渔业机械仪器研究所 | Cultured fish monitoring method and system based on image recognition |
CN112772485A (en) * | 2021-01-15 | 2021-05-11 | 广西壮族自治区水产科学研究院 | Ecological prawn breeding method |
CN113505649A (en) * | 2021-06-10 | 2021-10-15 | 广州杰赛科技股份有限公司 | Tap water chlorination control method and device |
CN113505649B (en) * | 2021-06-10 | 2023-11-17 | 广州杰赛科技股份有限公司 | Tap water chlorination control method and device |
CN114136906A (en) * | 2021-12-01 | 2022-03-04 | 浙江省海洋水产养殖研究所 | Intelligent fishery regulation and control method and system based on hyperspectral fish meat quality detection |
CN114620819A (en) * | 2022-03-01 | 2022-06-14 | 红云红河烟草(集团)有限责任公司 | Method for adjusting pH value of circulating water for spraying and washing cigarette peculiar smell gas |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110910067A (en) | Intelligent regulation and control method and system for live fish transportation water quality by combining deep learning and Q-learning | |
CN110728259B (en) | Chicken crowd heavy monitoring system based on depth image | |
CN109145032A (en) | A kind of bee raising intelligent monitoring method and system | |
CN109543679A (en) | A kind of dead fish recognition methods and early warning system based on depth convolutional neural networks | |
CN109583663B (en) | Night water dissolved oxygen amount prediction method suitable for aquaculture pond | |
Arvind et al. | Edge computing based smart aquaponics monitoring system using deep learning in IoT environment | |
CN113938503A (en) | Early warning system for diseases through live pig behavior sign monitoring and construction method | |
CN112506120A (en) | Wisdom fishery management system based on thing networking | |
CN106332855A (en) | Automatic early warning system for pests and diseases | |
CN102124964B (en) | Intelligent management system for mariculture | |
CN204653414U (en) | The assessment of a kind of Penaeus Vannmei factorial seedling growth Environmental security and prior-warning device | |
CN114037552B (en) | Method and system for polling physiological growth information of meat ducks | |
CN107830891B (en) | Data processing method based on aquaculture water pH value multi-parameter data acquisition device | |
CN110045771B (en) | Intelligent monitoring system for water quality of fishpond | |
CN113963298A (en) | Wild animal identification tracking and behavior detection system, method, equipment and storage medium based on computer vision | |
CN115604301A (en) | Planting environment monitoring system based on artificial intelligence | |
CN108897363A (en) | A kind of aquarium intelligence control system based on big data analysis | |
CN104007733B (en) | It is a kind of that the system and method being monitored is produced to intensive agriculture | |
CN116629550B (en) | Water environment supervision method and scheduling operation system based on cloud computing | |
Musa et al. | An intelligent plant dissease detection system for smart hydroponic using convolutional neural network | |
CN114898405A (en) | Portable broiler chicken abnormity monitoring system based on edge calculation | |
CN113326743A (en) | Fish shoal movement behavior parameter extraction and analysis method under breeding background condition | |
CN117029904A (en) | Intelligent cage-rearing poultry inspection system | |
CN115545962A (en) | Crop growth period control method and system based on multi-sensor system | |
CA3127938A1 (en) | Holding tank monitoring system based on wireless sensor network and monitoring method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200324 |