Case Study
ioCrops Labels 90,000 Instances
in Just 8 Days with Superb AI
ioCrops gathers the growth information of crops for automatic smart farming operations. However, they were struggling to find a scalable and automated way to collect and label this information.
Automatic data labeling through Superb Label’s Custom Auto-Label technology.
90% of raw and labeled data managed in one place
1.5x faster polygon segmentation with custom auto-label
Active learning workflows established
Reduced processing time through increased data quality
Successfully Commercializing an AI Model Using the Superb Platform
ioCrops is an Agtech company in Korea that remotely operates smart farms using their cloud data platform for agricultural data analysis, ioFarm, and their proprietary “monitoring robots.”
Smart farms refer to farms that are managed by remotely-controlled robots without any human intervention in the field. In such environments, damage to crops (paprika roots, stalks, etc.) caused by robots cannot be healed easily, unlike in traditional industries, so capturing the growth/reproduction information in near real-time and effectively managing robots are essential to smart farm operations.
Capturing and analyzing as much data as possible is key to solving operational issues and maintaining successful operations at smart farms. In this regard, ioCrops’ goal is to address the challenges of smart farm operations and gain valuable insights as quickly as possible with the help of technology.
“AI should be commercialized in a way that maximizes time savings in data training and processing for people. In that sense, the Superb AI Platform has done an excellent job by significantly reducing our data training time with its Custom Auto-Label feature.”
Yong-jin Shin, Machine Learning Engineer
ioCrops
Automating the Collection of Growth Data for Smart Farm Operations Proved Challenging
Unlike traditional farms, smart farms are dynamic environments that require increased attention to crop nutrition, growth, and reproduction. Tireless robots never cease to manage tomatoes, paprikas, and other farm crops to produce high-quality yields.
Traditionally, understanding the nutritional and reproductive growth of crops has been the realm of humans who examine stalks, leaves, fruits, flowers, and other elements of the crops. This task often requires human farmers to use tools such as vernier calipers (a ruler-like tool used to measure the length, height, and width of machinery or human body parts) to accurately measure the growth of crops throughout the day. However, this manual approach is heavily time-consuming and energy-draining, so some farms only sample a small portion of their crops or simply rely on their instinct and experience.
For smart farms to produce high-quality yields efficiently, accurate and objective growth data is a must. And ioCrops’ mission is to address this area with the most objective data possible.
In fact, ioCrops faced challenges right from the start in capturing diverse variables and image data from farms. Delays with data collection slowed down the overall speed of their pipeline. Given the nature of smart farms, there were clear limitations with automating the farms with human-collected data.
“We have prepared the three images below to demonstrate the challenges involved in collecting data from farms. The first image shows a person using a hand-held camera to capture photos, while the second image showcases a prototype robot that was deployed for autonomous data collection. Finally, the third image presents our nearly finalized robot, which is on the verge of commercialization. We had to proceed step by step as it was not possible to collect all the required data at once.”
Yong-jin Shin, Machine Learning Engineer
ioCrops
Data Labeling Automation Through the Superb Platform and ioFarm
To address these challenges, ioCrops developed a monitoring robot with automated driving and growth measurement capabilities. Using the robot, they created a suitable environment for automated data collection and leveraged the Superb Platform’s computer vision technology to extract growth metrics. The extracted data were then transferred to ioFarm and the Superb Platform for data processing and model training.
Significant Reduction in Data Training Time with the Help of Custom Auto-Label
プロジェクトのゴール
プロジェクトのゴール
ioCrops successfully reduced training data load by a considerable amount by adopting a Federated Learning approach, which involved having robots wander around the farm to collect data, sending that data to the Superb Platform, auto-labeling the data, and finally sending updated parameters for model re-training.
ioCrops needed to label their data quickly to expedite the development and advancement of their algorithm. With Superb AI's support, they significantly reduced labeling time per image, labeled 90,000 instances in just eight days, completed one entire iteration cycle in only 2~3 days, and ultimately succeeded in commercializing their AI.
How was ioCrops able to complete 1 iteration cycle in just 2~3 days?
ioCrops successfully reduced training data load by a considerable amount by adopting a Federated Learning approach, which involved having robots wander around the farm to collect data, sending that data to the Superb Platform, auto-labeling the data, and finally sending updated parameters for model re-training.
ioCrops needed to label their data quickly to expedite the development and advancement of their algorithm. With Superb AI's support, they significantly reduced labeling time per image, labeled 90,000 instances in just eight days, completed one entire iteration cycle in only 2~3 days, and ultimately succeeded in commercializing their AI.