Intelligent fish farm —the future of aquaculture

The developments of aquaculture engineering, mechanization and information technology and equipment construction are seriously lagging behind. Some breeding enterprises in underdeveloped areas are restricted by breeding technology and environmental conditions, breeding germplasm is degraded, enterprises lack good varieties, and aquatic products are of low quality. Aquaculture is facing huge challenges, but there is also a bigger opportunity. Ecological, facility, industrial, and intelligent are the future development directions of aquaculture.

Definition and system framework of intelligent fish farm

Intelligent fish farms rely on digital and intelligent technology to solve the problems of aquaculture labor shortage, water pollution, high risk and low efficiency. It can be divided into four categories according to different culture environments: pondtype intelligent fish farm, land-based factory type intelligent fish farm, cage-type intelligent fish farm and intelligent marine ranch.

Pond-type intelligent fish farm collects water quality information using sensors in real time, and unmanned aerial vehicle patrols to obtain the water surface activities of fish. Land based factory-type intelligent fish farm mainly realizes automated recirculating aquaculture (RAS). Cage-type intelligent fish farm obtains seawater quality and ocean current information using sensors and obtains fish movement and feeding information using machine vision and sonar.

Intelligent marine ranches usually use high-definition surface cameras and underwater robots to collect the video information of the ranch in real time, transmit the video to the data server in the shore-based information control center for the purpose of biological identification, behavior analysis, and biomass estimation.

Advance information technology in intelligent fish farm

The traditional aquaculture IoT system adopts three-tier structure. The device layer is composed of sensing equipment, control equipment and data acquisition terminal. The sensing equipment is responsible for collecting the environmental data as well as the working status of the device and aquaculture video image information.

The control equipment includes aerator (oxygen cone), feeder, pump valve and other aquaculture equipment. The data acquisition terminal is responsible for the upward transmission of sensor data and the reception of control instructions. The network layer generally adopts wireless network. The cloud service layer includes cloud platform and smartphone app.

For businesses with high real-time requirements, end-to-end millisecondlevel low latency is required, but it is difficult for ordinary cloud computing models to meet the above tasks.

Therefore, this research introduces Edge computing and 5G into the aquaculture IoT system to improve the standardization, stability and usability of the system.

The work of fishery informatization will produce a large number of multidimensional data. By simulating human thinking and intelligent behavior, AI can learn the massive information provided by the IoT and big data, analyze and judge the problems, finally complete the decision-making task, and realize the accurate operation of the fish farm.

Intelligent hardware for measure, control, and self-feeding

Water environment ecological monitoring refers to the use of sensors and cameras carried by unmanned ships or surface buoys to automatically collect water quality parameters, aquaculture biological pictures and video information, and then store, transmit, analyze and predict data. Long time accurate detection of aquaculture water quality parameters provides a reliable data source for automatic control and intelligent decision-making of intelligent fish farm.

Intelligent aeration system refers to the equipment that can accurately measure and control the DO in water, which is composed of various sensors, network transmission module and IOT actuator. The intelligent aerator can monitor water temperature, air humidity, air pressure and DO in real time.

Automatic feeding system has been widely used in industrial recirculating aquaculture, including automatic feeding system with multi monomer centralized control and automatic feeding robot system. In the pond-type intelligent fish farm, intelligent bait-dropping equipment should be deployed on unmanned boats or UAV, and the UAV will be responsible for the independent transportation and loading of bait.

Unmanned patrol system

Biomimetic robot fish carries a variety of sensors to automatically monitor the water quality and the operation status of key equipment.

Biomimetic robot fish can also monitor the feeding rule of fish based on computer vision technology and analyze relevant data to provide a basis for optimizing the feeding strategy.

The inspection robot based on the deep learning, computer vision and positioning technology can detect the position of sick and dead fish and use the automatic manipulator to pick up the dead fish combined with the optical and acoustic system.

Underwater robot in cage culture can also locate the damaged and contaminated positions of the net clothes and use tool to clean and repair the net clothes. Orbital robot can inspect the circulating water pipe network, oxygenation equipment and feeding equipment in the recirculating aquaculture workshop according to the predetermined inspection route.

Intelligent harvesting system

Intelligent harvesting system is the last module for fish farm to complete the breeding cycle. Using this system, the breeding objects will enter the market through the transportation with or without water. At present, trawling is the most efficient way of fishing.

Water quality soft measurement and control method

Aquaculture water quality greatly affects the growth rate, health status and feed intake efficiency of fish. A single water quality parameter in aquaculture water is easily affected by other water quality parameters, which increases the difficulty of detection by a single detection method, and also provides the possibility for the application of soft measurement.

The basic idea of soft measurement technology is to infer or estimate important parameters that are difficult to observe with the help of some easily observed variables. If a sensor fails, it will directly lead to the deviation or even failure of the soft measurement prediction.

Intelligent feeding strategy

Feeding back the evaluation results of bait-feeding effect to the control system will help to adjust the feeding amount in real time. The artificial experience model is usually based on a large amount of observation experience in aquaculture, and a regression fitting analysis method is used to establish a mathematical equation related to the nutrient requirements of fish growth and the amount of feed.

Behavior analysis of raised species

Fish bioassay is one of the earliest biological monitoring methods, especially used to investigate the impact of pollution fluctuation on fish behavior. Up to now, the indicators used to monitor and evaluate water quality mainly include movement behavior, respiratory behavior, and group behavior.

Using computer vision technology to monitor fish behavior and obtain fish hunger levels can improve ability and accuracy of image processing and provide a theoretical basis for intelligent feeding.

Deep learning is the most advanced machine learning method at present. It comes from the research of artificial neural network architecture which has a large number of hidden layers and millions of parameters. Intelligent fish farm tries to combine machine vision, sonar detection and deep learning technology to realize behavior analysis of cultured animals in real time.

Biomass statistics

The biomass statistics is crucial to support the fish farmers’ decisions such as fish food dosage, drug consumption and fish loss. The length, width, area, and circumference of fish in different growth periods are closely related to their weight. These parameters will be used as an important basis for the estimation of fish biomass. Laser scanning technology is another noninvasive monitoring technology that can be used to estimate fish biomass in real time.

However, compared with the limited penetration ability of light in water, the attenuation of sound wave in water is much smaller. Identification sonar is a kind of multi-beam system which uses acoustic lens to transmit independent beam.

Modern advanced remote sensing technology using satellite remote sensing images and relevant professional software to analyze the marine fishery resources can accurately obtain the position of fish stock, which greatly improves the accuracy and quantity of fishing.

Diagnosis of fish diseases

After a healthy fish gets sick, it is usually accompanied by changes in the color and texture of the body surface.

Since different fish suffer from the same disease with different symptoms, the first step in studying fish diseases is to identify the type of fish by analyzing the sub-images of the sick fish body and extracting its features including the color feature and texture feature based on the statistical method and the wavelet method.

Current research on fish images are limited to obtaining excellent recognition and detection results under certain conditions. To improve the accuracy and sensitivity of the automatic fish disease diagnosis system in the intelligent fish farm, it is an effective way to add water quality analysis, fish behavior analysis and meteorological data analysis as the correction input of the deep learning method.

Fault diagnosis of equipment

An intelligent fault diagnosis process is divided into two steps. Firstly, all data need to be preprocessed, the feature parameters which can represent the fault symptoms are extracted based on deep learning, and a certain number of sample sets are selected to train the neural network to get the expected diagnosis network and classifier. Secondly, according to the trained neural network and classifier, the online data of the system is diagnosed.

Challenges

Aquaculture robots are facing great challenges in the aspects of optimization of target identification and positioning algorithm, optimization of navigation and path planning algorithm and optimization of operation object sorting and monitoring algorithm.

Compared with traditional machine learning, deep learning can better extract the features of agricultural images and structured data, and effectively combine with agricultural machinery to better support the development of aquaculture intelligent equipment. It is found that there are still some deficiencies in the application of deep learning in aquaculture as follows: Firstly, deep learning needs large data sets for model training, verification and testing. Secondly, most of the aquaculture problems based on deep learning are supervised learning, and the corresponding sample data need to be labeled.

Conclusions

Modern emerging technologies such as AI, big data, IoT, sensors, machine vision and robots will gradually participate in the whole process of aquaculture production for liberating traditional labor force completely, and finally realize multi-scene all-weather real-time monitoring of production environment, big data analysis based on cloud platform and real-time intelligent decision-making.

The construction of intelligent fish farm is much more complex than other intelligent farm projects. The reliability and service life of sensors, the robustness and accuracy of analysis and decision-making models, the reliability of data transmission based on IoT technology, and the cooperation efficiency among various aquaculture intelligent equipment also need to be further solved.

This is a summarized version developed by the editorial team of Aquaculture Magazine based on the review article titled “INTELLIGENT FISH FARM—THE FUTURE OF AQUACULTURE” developed by: CONG WANG, ZHEN LI, TAN WANG, XIANBAO XU, XIAOSHUAN ZHANG, AND DAOLIANG LI. The original article was published on SEPTEMBER 2021, through AQUACULTURE INTERNATIONAL under the use of a creative commons open access license. The full version can be accessed freely online through this link: https://doi.org/10.1007/s10499-021-00773-8