Alongside the shrimp ponds in Kaohsiung, Taiwan, two professors from National Sun Yat-Sen University (NSYSU) — Hung Chin-Chang from the Department of Oceanography and Huang Ing-Jer from the Department of Computer Science and Engineering, along with their research team invited the guests to try some lightly boiled shrimp. Holding the boiled shrimps with rich color and firm texture, the two professors beamed with confidence as they announced, “Our AI shrimps are farmed using Artificial Intelligence, and they taste absolutely amazing!”
To help shrimp farmers improve their productivity and the quality of their harvests, the research team is using artificial intelligence and big data analysis on its fast-growing data set.
According to the recent study conducted by Seagate, Data Pulse: Maximizing the Potential of Artificial Intelligence, there has been robust adoption of AI technologies in the Asia Pacific region. The study showed that 62% of respondents in Taiwan have implemented AI in one or more areas of their business, such as Supply Chain / Logistics (73%), IT (58%), Product Innovation / R&D (56%) and Customer Support (50%). What’s more exciting is that now the use of AI is making inroads into the traditional aquaculture industry in Taiwan.
Professor Hung has been deeply concerned about the wellbeing of the people who depend on fishing to make a living, and has been communicating regularly with breeders, producers and sales personnel from the southern Taiwan every year. He recalls that every time he goes to the fishing village, the most commonly asked questions are, “Professor, how can I know if the fish and shrimps are full? How can I tell if they are healthy or not?”
The critical challenge: water quality
The most critical challenge for the farmers lies in the poor water quality of their ponds. Shrimp growth is highly dependent on the right amount and timing of feeding, but the turbid water makes it almost impossible to tell with the naked eye if there is any leftover shrimp feed in the water, or if the shrimps are healthy. Hence farmers may have the tendency to overfeed, which then leads to excessive leftovers and waste, making the water even more turbid. To make things worse, the excess feed can eventually become a substantial nutrient source for bacteria. Some of these bacteria may be pathogenic and end up killing the whole pond of shrimps.
In the old days, farmers had to rely on their own experiences to take the right action, but chances are they might often be one step behind — by the time they realize they’ve over-fed, or that the shrimps are sick, it’s usually too late. “If there is one dead shrimp floating on top of the ponds, oftentimes there are hundreds of others that may have already died at the bottom,” Hung said sympathetically. “Especially when typhoons or extreme cold weather strike the areas, these farmers will suffer from tremendous financial loss.”
There are hundreds of solutions to this problem, and the easiest way is to monitor the concentration of nitrous acid in the water regularly, check the remaining amount of feed, and judge the shrimp’s health based on their appetite. But none of these solutions is simple for farmers using traditional methods that have no mechanism enable timely and comprehensive assessments of the underwater situation.
“I asked myself, is there a way to adopt scientific management into the traditional aquaculture business?” Moved by the plight of fish farmers, many who continue working into advanced age, forced to attend to their breeding pools even in the midst of inclement weather, Hung decided to join forces with Professor Huang to form the AI Aquaculture Research team. With one an expert in shrimp farming, and the other in computer science, the two professors focused their expertise on finding solutions to improve the processes used by shrimp farmers — reducing their workload, increasing their revenue, and offering new hope for revitalizing the fishing village.
A new kind of underwater camera as the eye; AI as the brain
In 2016, Hung Chin-Chang collaborated with Taiwan Ocean Research Institute (TORI) to develop a highly sensitive underwater video system (UVS). Their work was discussed in the scientific journal Scientific Reports and triggered discussions among aquaculturists, contributing to further academic and practical explorations.
According to Hung, underwater cameras on the market are not suitable for the use in turbid aquaculture ponds. This particular underwater video-enabled aquaculture system, on the other hand, can not only record full-color real-time footage, but can also allow observation of the fish and shrimp all day long with infrared technology. It also allows the farmers to remotely control the feed dispenser and sludge disposal system through a mobile app or the Internet, which has reduced physical labor hours and minimized the danger of working at night.
However, video surveillance alone is not good enough. “Traditional shrimp farming depends heavily on nature, so finding a regular pattern of feeding is extremely difficult. That is the biggest obstacle to implementing automated management system in aquaculture,” said Huang, “But AI can fill the gap.”
The AI Aquaculture Research team analyzes large amounts of data, including the enhanced surveillance images, to “train” the machines to make decisions about the feeding schedule. With the machine learning technology, the team is also able to build a future shrimp growth model, based on the number, size and moving speed of shrimps in the ponds, for better results in maintaining healthy population growth. Further, the technology can turn the shrimp farmers’ decades’ worth of experience into digital data that can be passed down to successive generations.
Data analytics used to extend efficiency in shrimp farmingFor example, when the intestines of the shrimp in the image appear as broken lines or even disappear, it usually indicates the shrimp is sick or dead. AI may help the farmers to take preventive action by disposing of diseased individual shrimps before the disease has a chance to spread.
Hung said, “Traditional feeding machines only act on whether or not to dispense feed, but farmers still need to adjust the frequency and quantity of feeding manually. Since feed accounts for about 40% of the overall costs, keeping the feeding amount precise is actually saving money. Furthermore, once we collect enough data to build the growth model for shrimps, farmers will be able to determine when’s the best time to sell their shrimp based on scientific data. We hope that by adopting AI in the traditional aquaculture industry, the overall output value and benefit could be significantly improved.”
“This is a critical example of how organizations in various industries can benefit from data and AI technology and improve people’s lives and their livelihoods, revitalizing traditional methods of shrimp farming without losing the deep experience and knowledge that farmers have cultivated over the years,” said Robert Yang, vice president, Asia Pacific sales of Seagate Technology. “The AI Aquaculture Research project team led by Professor Hung and Huang leverages advanced technology and massive data to solve the farmers’ problems and improve the productivity. Our Data Pulse study indicates a fairly high percentage of Taiwanese organizations have started exploring AI, as they see its potential to grow their businesses in the digital transformation.”
“Our team’s data volume increased at least 10-fold. The raw data generated in merely one year reached a staggering 47 TB,” said Huang. “Adding on the 4,000 images transferred to the AI database daily, it is estimated that over the course of three years, the data generation of our team will be close to 500TB. All of this data collected under different conditions provides valuable insights for both biological research and the aquaculture industry. The data deserves to be stored safely for good.”
IT infrastructure is the key to building AI database
Seagate’s “Data Pulse: Maximizing the Potential of AI” study showed that 76% of respondents had planned to adopt or start to adopt more AI solutions, but 83% of respondents said they have no idea where to start.
“Most of the organizations here in Taiwan have not yet built their own AI database, especially those in traditional industries where they still rely on past experience to make decisions on site,” Huang shared. “In order to realize digital transformation, these organizations must discover the hidden value of their data, where building IT infrastructure is the key.”
Hung believes that the most ideal system for AI aquaculture would require 20 sets of underwater photography modules operating synchronously — and the amount of raw data generated each year would reach 656TB. With an explosive increase in data volume, building an expandable IT infrastructure will be critical.
To cope with the surge in data volume, the NSYSU AI Aquaculture Research Team is now storing their video surveillance data using Seagate SkyHawk AI hard drives, which are designed specifically for AI-enabled video capture and support up to 64 HD cameras and 16 additional AI analytics streams. In addition, to build their AI database, Seagate IronWolf Pro NAS hard drives are installed to meet the mass data storage demand. Seagate Backup Plus Ultra Slim and other external drives are also used for data backups in case of internet disconnections caused by natural disasters.
There are local aquaculturists now discussing about transferring this “AI Shrimp” farming to their own ponds. “Taiwan’s agriculture and fisheries are choosing to scale down and specialize in select items. If we can make good use of the power of science and technology to improve the efficiency and increase harvest results through AI, this industry will have the potential to be revitalized,” said Hung.