Follow alphago strategy in upgrading the latest cu

Follow the alphago strategy in upgrading the New Cucumber sorter

Abstract: the application of artificial intelligence in the field of agriculture has just begun, and the challenge is greater than any other industry, because agriculture involves too many unknowable factors. Geographical location, surrounding environment, climate, water and soil, diseases and pests, biodiversity, complex microbial environment and so on, these factors are affecting agricultural production

it is reported that Makoto Koike, an engineer, began to develop a new method of sorting cucumbers, built a cucumber sorter, and used the tensorflow deep learning software framework released by Google to the public in 2015. Moreover, he also established an automatic conveyor system to transfer each cucumber from the photo station to the box designated by the program

Makoto Koike's mother has been sorting cucumbers manually for decades. Today, Makoto Koike is trying to "train" a machine to replace it

mako oil pump starts working to Koike is an engineer. For a long time, he has been interested in repairing electronic accessories and machines; It can be said that he is not a natural person who likes natural outdoor life. In 2014, when Makoto Koike was 33 years old, he left the city where he had worked and lived for a long time and moved to Shizuoka Prefecture, a beautiful city, to help his parents run a cucumber farm. "I think I have reached a certain age," Makoto Koike said. "I want to be closer to my home and family."

makoto Koike's family often give wholesalers the better, straight and uniform thickness cucumbers, while the remaining imperfect cucumbers are sold at half price. Makoto Koike's mother has been using reassurance for a long time. Now let me briefly tell you a few points, sorting these vegetables one by one and putting them into different boxes by categories. Although each cucumber took her only half a second, this work occupied most of her working time; Sometimes, she even handled more than 4000 cucumbers a day in a few days

makoto Koike believes that classifying cucumbers should not be the main work of melon farmers. The most important task of melon farmers should be to focus on planting delicious cucumbers. So he decided to entrust the classification work to the machine, but the cucumber classifier on the market is either poor performance or too expensive, which is not suitable for small farms. Last spring, Makoto Koike began to develop a new method of sorting cucumbers. He built a cucumber sorter and used the tensorflow deep learning software framework released by Google to the public in 2015

part of Makoto Koike's inspiration is an article he read about alphago, the first computer program to defeat a human go master in history. In the case of alphago, it extracted 30million pictures from the real go game to help determine which action step is the most effective. Makoto Koike also hopes to create a similar strategy to help it sort cucumbers

advanced AI technology including in-depth learning belongs to the field of professional researchers and software companies. However, recently, some technology giants, including Google, Facebook, Microsoft, Amazon, Baidu and universities, have released free open-source versions of tools, so that non professional programmers such as Makoto Koike can also access them

in his project, raspberry Pie 3 is used as the main controller, and a custom photo shooting station is established, which enables him to shoot each cucumber from three different angles. Then, in order to analyze these images, they were sent to the tensorflow platform. At first, they ran on a small neural network to judge whether it was a cucumber. Then, the photos that had been determined to be cucumbers were transmitted to a larger neural network based on Linux server to classify cucumbers according to different characteristics

however, before he can really use AI technology to sort cucumbers, Makoto Koike must first "train" this system. In order to train this model, Makoto Koike spent three months "feeding" it 7000 cucumbers photos, which are classified and labeled by Makoto Koike's mother. Finally, he also established an automatic conveyor system to transfer each cucumber from the photo station to the box designated by the program

makoto Koike completed the development of this artificial intelligence system last year, and to some extent, it really worked. However, its classification accuracy of cucumbers can only reach 70%, which is too low. They must also conduct manual inspection. Moreover, at present, these vegetables need to be placed on the photo station one by one, that is to say, Makoto Koike's mother has not been "completely replaced"

the R & D and application of artificial intelligence in agriculture have begun as early as this century, including intelligent robots such as farming, sowing and picking, intelligent identification systems such as intelligent soil detection, pest detection, climate disaster warning, and intelligent wear products for livestock used in livestock breeding

the application of artificial intelligence in the field of agriculture has just begun, and it faces greater challenges than any other industry, because agriculture involves too many unknowable factors. Geographical location, surrounding environment, climate, water and soil, diseases and pests, biodiversity, complex microbial environment and so on, these factors are affecting agricultural production. If you test a successful algorithm in a specific environment, another environment may not be useful

most of the successful applications of artificial intelligence we have seen at this stage are in specific geographical environments or specific planting and breeding modes. When the external environment changes, how to challenge algorithms and models is a challenge for these AI companies, which requires more collaboration between industries and agronomists