The AI system recognizes durian images and counts the number of fruits. [Photo/en.xmu.edu.cn]
From May to July, durian orchards in Malaysia, the fruit's primary growing region, are bustling with activity. In the orchards, farmers are busy taking photos of the trees — not for documentation, but to count the durians. Thanks to an AI system developed by the School of Artificial Intelligence and Robotics (SAIR) at Xiamen University Malaysia (XMUM), this process now takes just five seconds per tree, compared to the 300 seconds it used to take manually.
This AI technology marks a significant breakthrough in the durian industry, addressing a longstanding challenge in predicting durian yields. Traditionally, farmers relied on manual counting, which was inefficient and often inaccurate due to the height and dense foliage of durian trees. The AI system, however, can swiftly and accurately count the fruit, even when up to 50 percent are obscured by leaves, with a 95 percent accuracy rate.
Led by Professor Zhang Yingqian, the team spent six months developing a customized AI vision system. They collected 1,500 images under various lighting and viewing angles and expanded them to 5,000 samples through data augmentation. By combining spatial attention mechanisms and feature pyramid technology, the system can identify fruit even when partially obscured.
Designed for practical use, each tree has a unique QR code, and the lightweight model runs on cloud servers compatible with various smartphones. Workers capture images and upload them via a local network for instant analysis.
Beyond harvest forecasting, the team envisions a fully AI-powered durian supply chain — from precision irrigation to product development — and hopes to eventually bring the technology to other markets, including China.
The durian processing workshop. [Photo/en.xmu.edu.cn]
