The era of AI smart agriculture is coming, and whether enough data can be collected to train robots to solve the problem of automated harvesting will be the key to digital transformation.
According to the "World Agrochemical Network" report, the application-based "smart agriculture" market is expected to reach a scale of US$18.45 billion in 2022, with a compound annual growth rate of 13.8%. Smart agriculture has a wide range of applications, such as remote data monitoring of sunlight, temperature and humidity in farms, crop growth monitoring, fruit picking robots, even pest control, and regional 3D vegetation detection. The scope of smart agriculture.
Facing The Trend of Smart Agriculture, Countries Around the World Offer "Data" To Solve the Problem
France, as the EU's largest agricultural producer, bears the brunt of this. The French government, agricultural organizations, and private enterprises have collaborated to establish an agricultural information database covering planting, fishing, animal husbandry, and even agricultural technology research and development, commercial markets, and legal policies. French farmers don't have to go out under the sun, just swipe their mobile phones, and they can master the world's "farming affairs" with one hand.
In Asia, Japan is notoriously aging country, with the average age of farmers as high as 67 years old. The Ministry of Agriculture, Forestry and Fisheries of Japan estimated that in 2015, there were still 1.5 million agricultural employees, but by 2030, it will drop all the way to 750,000, and it will be reduced by half within 15 years. This figure made the Ibaraki prefectural government decide to take action to save local agriculture.
Ibaraki Prefecture is located in the northeastern part of Japan, with a vast area of agricultural land, about the size of 460 Tokyo Domes. In April of this year, Ibaraki Prefecture launched the "Tsukuba City Future Co-creation Project". Through industry-government-academia cooperation between the government, farmers and new start-ups, it will jointly develop AI robots for smart agriculture with farmers, and introduce robots in a low-cost way. Locally grown tomato, cucumber, green pepper, lychee and other farms. Committed to creating time-saving, labor-saving and "earned-money" agriculture with AI video surveillance, creating the prosperity of sustainable agriculture in the next 100 years.
Smart Agriculture Starts with Data Collection
The smart agriculture projects I have handled in the past have covered main types such as automatic harvesting, growth monitoring, and pest control. Some of the clients are from large enterprises, start-ups or agricultural organizations, among which Japan has the most clients.
Before embarking on digital transformation or AI industrialization, "data collection" is an important way to start. For example, we all know that a top-quality Wagyu beef needs to have fine meat quality and evenly distributed oil flowers, and a good data set is very similar to Wagyu beef and needs to have the following characteristics at the same time:
- Is the “quality” of the images clear and correct?
- Are the “proportions” of various target objects and field situations evenly distributed?
If a batch of AI data is collected from the wrong angle, or if there is a large deviation in the number of images among various target objects, and the images are blurred, it is easy to cause machine learning errors.
Taking the AI application of the growth monitoring of flower farmland as an example, a high-angle aerial camera is generally used for framing, and when collecting pictures, the surrounding objects that will cause interference, such as weeds, vegetation, other mixed objects, are not overlooked. Flower varieties, etc. If necessary, even the influence of sunny and rainy days should be taken into consideration.
The AI application of fruit harvesting robots needs to be photographed from a head-up angle, and may focus on obtaining features such as stems, leaves, buds, flowers, and fruits that are similar in proportion and clear, in order to learn quickly and well. .
After the data collection is completed, it enters another important link - data annotation.
How To Label Ai Data Is the Key
The AI data labeling of smart agriculture is actually not simpler than the general data labeling, because it involves a lot of "botany" and pays great attention to details.
For a while, we had an entire row of potted tomatoes in the office. After a while, tomatoes were replaced with raspberries, lilies and other plants. Some customers come to visit, thinking that these potted plants exist to beautify the environment, but in fact, the project management and AI data annotation team can grasp the characteristics and details of the annotation in order to observe the details of plant growth up close. When encountering problems that cannot be solved, we will also consult experts related to agriculture, so as to fully understand the growth characteristics of plants, so that the labeling work can be done in place and with high quality.
Because there is abundant data labeling experience as a nutrient, when dealing with different agricultural projects, past experience can be quickly copied to other project types. Take the case of the raspberry harvesting robot, for example.
If only the fruit and flower parts are framed when labeling, and machine learning is performed, a blind spot will be found: that is, the "branches" also need to be labeled, and the difference between the trunk and the branches must be clearly distinguished. Why do you say that?
If only the fruit is marked, the machine may directly cut off the branches in order to achieve the goal of harvesting the fruit, causing serious agricultural damage. At the same time, the branches are divided into two types: the main trunk and the branch trunk. The machine must be clearly told several principles that do not contradict each other, so that the machine can know which parts should be cut and which ones must not be cut.
The reason sounds simple, but from the perspective of customers and AI robots, we need professional insights into data and innovative solutions. Through the feedback mechanism, it not only optimizes the labeling principle, but also saves the time for customers to correct data errors back and forth.
The era of AI smart agriculture is coming. Under the promotion of production, government, and education, with the assistance of AI data, the work of harvesting rice in the future may be easier, and may no longer be hard work.
