To solve the pain points of the aquaculture industry, people first land, and artificial intelligence can land Fish farming is a high-labor, high-risk and high-tech-intensive industry. Although the profit is not low, it also faces high risks such as natural disasters and fish diseases. As long as there is a slight avoidance and no attention is paid, it may cause millions of fishermen Losses, such as the sudden mass death of Yilan golden pompano before, the risk of rapid temperature changes, etc.; in addition to high risks, the breeding industry also has a lot of technical experience that cannot be humane to outsiders but is also difficult to pass on. It cannot be seen, the quantity cannot be controlled, and these factors have increased the entry threshold of the aquaculture industry, and have also made many farmers accustomed to spending time on patrolling fish farms, fearing that it will be too late to rescue the fish farms when they are in trouble. However, through the introduction of technology, fish farmers can now keep abreast of the growth status of farmed fish and shrimp underwater through their mobile phones, thereby reducing unnecessary labor costs. How did Haisheng Technology, which developed this underwater monitoring system, turn these experiences that are difficult to digitize into data that can be used by computers? How about automation or even combining advanced technologies such as artificial intelligence, Internet of Things, and mechanism design to develop a breeding system? For fish farmers, because they cannot see the fish, they cannot tell whether the fish is sick or dead, unless they are picked up for inspection. "However, such behavior may also cause the death of the fish," said Lian Weizheng, co-founder of Haisheng Technology The doctor said that dead fish may not float, but they will carry viruses or bacteria. Once other fish are infected and spread to the whole pond, the loss will be as high as millions. "Fishermen have no chance to struggle. If the capital is not enough thick, that is direct bankruptcy.” Underwater image processing technology enhances fish identification In order to solve the pain points that fishermen cannot see, measure, or control, Haisheng Technology has developed a monitoring system for raising good fish and aquatic products that combines artificial intelligence, Internet of Things, mechanism design and other technologies. Automatically collect fish data and send it to the IoT AI field control analyzer in the cloud, analyze the fish through the built-in artificial intelligence search engine and automatic image enhancement in the cloud, and then integrate field devices through the self-built API docking platform, such as water quality detection This system has been applied to eight species of high economic value fish such as clownfish, sea bream and sea grouper. The most important thing here is the underwater image processing. Lian Weizheng said that although the underwater monitoring system in the past recorded a lot of images, more than 70% of the images were either not recorded, or the fish could not be seen clearly. " Fishermen need to spend 20 to 30 minutes to find a video with fish traces, because the efficiency is too low and the willingness to use it is low." So they proposed a cloud-based intelligent breeding fish school time-short-time search and image enhancement system as a solution, which is applied to object recognition , tracking, automatic grouping and other artificial intelligence technologies filter out images worthy of viewing by fishermen, and fully automatic underwater image defogging technology enhances the recognition of underwater images. Since the underwater environment is more complicated than that on land, in addition to the change of light refraction, there is also the interference of fogging. Therefore, the Haisheng Technology team proposed a set of fully automatic intelligent underwater image defogging algorithms, which can fully automatically analyze and adapt to various impacts. gender correction. "Because the images shown to AI are different from those seen by humans, how to make an algorithm that can be accurately identified by AI and clearly seen by humans is the goal of the team's efforts," said Weizheng Lian.
Forum Role: Participant
Topics Started: 0
Replies Created: 0