China's first big fishing model "Fan Li Big Model 1.0" comes out, will change the future of fishing

From: National Digital Fisheries Center

        On June 15, China's first big fishery model "Fan Li Big Model 1.0" was released in China Agricultural University, it is reported that the model can achieve fishery multi-modal data acquisition, cleaning, extraction and integration, etc., will provide comprehensive and accurate intelligent support for fishery workers, management operators and government decision-making departments.

"Fan Li big model 1.0" release site.

Big fishing country, facing the needs of transformation
        China is a big country in aquaculture, data show that in 2023, China's aquaculture production reached 58.12 million tons, accounting for about 60% of the world's total aquaculture production, providing 1/3 of urban and rural residents with high-quality animal protein. But at the same time, China is not a strong country in aquaculture, aquaculture resource utilization, labor productivity is low, and the development of aquaculture industry is facing a variety of transformation needs. Fan Lida model designer, initiator, director of the National Digital Fishery Innovation Center, Professor Li Daoliang of the College of Information and Electrical Engineering of China Agricultural University introduced, "There are many varieties of aquaculture in China, including fish, shrimp, crab, shellfish, ginseng, algae, etc., and diverse farming modes, so it is extremely difficult to establish a complete production model of aquaculture varieties; At the same time, the labor force has a general aging phenomenon, survey data show that in China's aquaculture, labor costs accounted for about 70%, the average age of workers reached 55 years old. The new generation lacks farming experience and is not willing to engage in traditional farming production, and needs the support of AI technology."

Fan Lida model designer, initiator, director of the National Digital Fishery Innovation Center, Professor Li Daoliang, College of Information and Electrical Engineering, China Agricultural University.

        With the development of modern technology, aquaculture has developed from the 1.0 era to the 4.0 era. Li Daoliang introduced, "The 1.0 era of fisheries is mainly based on small-scale farmer production, characterized by relying on manpower, hand tools, experience and other farming. In the 2.0 era, aquaculture has gradually realized mechanization and equipment, mainly relying on mechanical power and electricity for production. In the 3.0 era, automation and computer technology have become the core, and production equipment has the characteristics of digitalization, networking and automation. In the 4.0 era, technologies such as the Internet of Things, big data, artificial intelligence and robots are widely used in production, and unmanned production is gradually realized."
        With the gradual emergence and maturity of artificial intelligence, robot learning and other technologies, more and more agricultural scenarios have begun to apply these technologies, but as a large country in aquaculture, China's current aquaculture, the application of relevant technologies is still relatively lacking.

Fishery model, upgrading from small to large
        How can modern technology be applied to aquaculture and even create unmanned fisheries of the future? Li Daoliang introduced that there are many varieties of aquaculture in China, and the breeding environment is very different, and the construction of the mechanism model needs to consider a series of factors such as fish species, bait, diseases, and environmental changes. In the face of numerous varieties and farming modes as well as regional climate differences, it is unrealistic to establish a breeding mechanism model like that of developed countries one by one.
        The so-called large model refers to a machine learning model with large-scale parameters and complex computational structures, and the number of parameters is often billions or even hundreds of billions. In fisheries, large models can leverage deep learning and data-driven approaches to analyze vast amounts of farming data to reveal patterns and correlations. "They can not only simulate and predict the impact of water quality, food, disease and other factors on the breeding results, but also optimize the breeding plan to improve production efficiency and economic benefits." Li Daoliang said.

Smart pond fish farming scene

        With the development of society and the transformation of aquaculture industry, the large fishery model has become an important support for industrial development. To this end, Li Daoliang led the team to cooperate with the three operators of China Unicom, China Telecom and China Mobile, as well as major aquaculture colleges and research institutions in the country, focusing on 27 kinds of aquaculture text corpus such as fish, shrimp, crab and shellfish in China. Supplemented by multi-modal data such as text, image, video and audio, a large-scale corpus of fishery expertise is formed. Through deep learning architecture, pre-training and fine-tuning, parameter sharing and attention mechanism, prompt engineering and other technologies, multi-modal fishery data collection, cleaning, extraction and integration are realized. "This model not only realizes rich knowledge generation of fishery culture, but also includes diversified prediction, analysis and decision-making of water, bait, disease and tube." Li Daoliang said.

Fan Li in the name of changing the future of fisheries
        After the successful construction of the large model, it was named "Fan Li Large Model 1.0". Li Daoliang introduced, Fan Li is the late spring and Autumn state of Yue doctor, as we all know, he is a famous politician, military, is also the originator of businesses, but he is also the earliest aquaculture experts in our country, as early as 2500 years ago in the Spring and Autumn period, he wrote a "fish culture", and spread to the present, "so we take the name of Fan Li, hope it can be in the new era, A new climate for aquaculture in China."

        According to reports, Fan Li Da model 1.0 is divided into ask me, please listen to me, look at me, please make four modules, respectively representing text, voice, video, Internet of things decision four scenarios, users can query different applications of fisheries. In order to accurately monitor and evaluate the health status and abnormal weight of fish, which is time-consuming and laborious, and may cause harm to fish, the National Digital Fisheries Innovation Center has developed a fish weight estimation model based on computer vision technology, and a deep neural network algorithm based on real-time capture of underwater fish images by machine vision. Automatically complete the detection and positioning of fish targets in images, and realize real-time and accurate estimation of fish weight in a non-contact way by extracting multi-dimensional features such as shape, color and texture, synchronously complete the monitoring and calculation of growth and health status, providing data support for feeding decisions, water environment and energy consumption optimization control.        

Fan Li Da model utilizes a variety of modern technologies to realize the digitalization and unmanned aquaculture. The picture shows the species recognition model of fish.

        "At present, the Fan Li model is still 1.0, and the future will continue to evolve, and the application of artificial intelligence in smart fisheries is diversified and far-reaching, long-term, and cannot be achieved overnight." In the future, Fan Li grand model still has a long way to go, we must give full play to the advantages of communication, scientific research, aquaculture enterprises, farmers and other different fields, to promote the development and application of large models in collaboration with production, education and research, artificial intelligence can really land." Li Daoliang said.