As a staple food for more than half of the global population, rice yield security has always been a top priority in the agricultural sector. However, weeds in rice fields—especially grassy weeds such as Echinochloa crus-galli (barnyard grass) and Cyperus difformis (smallflower umbrella sedge)—bear striking morphological similarities to rice seedlings in the early growth stage. Competing for light, nutrients, and water, these weeds can cause yield losses ranging from 20% to 100% in severe cases. Traditional manual weeding is not only time-consuming and labor-intensive, but also unable to meet the needs of large-scale cultivation. Blind use of herbicides, on the other hand, may pollute soil and water sources, posing threats to the ecological environment. How to accurately identify and targetedly eliminate weeds at their early growth stage has become a major challenge in agricultural production.
In recent years, with the development of remote sensing technology, hyperspectral imaging has emerged as a new solution to this problem. Equipped on unmanned aerial vehicles (UAVs) or ground-based sensors, special cameras of this technology capture hundreds of narrow-band spectral information of plants across the visible to infrared range, forming a spectral signature. Just as each person has a unique fingerprint, each plant species reflects a distinct light signal due to differences in leaf structure, chlorophyll content, and other traits. For instance, healthy rice exhibits strong reflectance in the green light band (545–565 nm), while Echinochloa crus-galli shows significant differences in reflectance from rice in the near-infrared region (700–1350 nm). By analyzing these signatures, technicians can distinguish weeds from rice as soon as the weeds emerge.
A review by Professor Abdul Shukor JURAIMI’s team from Universiti Putra Malaysia points out that hyperspectral imaging technology boasts advantages of non-contact operation, high precision, and early detection. Compared with traditional manual visual inspection, it can complete detection within 10–30 days after rice sowing—a critical period when weeds are most competitive—with an identification accuracy generally exceeding 90%. For example, regarding Echinochloa crus-galli and weedy rice (Oryza sativa f. spontanea), the most common weeds in rice fields, researchers achieved identification accuracies of 100% and 92%, respectively, by analyzing spectral data with intelligent algorithms. This accurate identification lays the foundation for targeted weeding: combined with UAVs and prescription mapping technology, it enables site-specific herbicide application, reducing pesticide usage by up to 50%. This not only cuts costs but also alleviates environmental burdens. The relevant article has been published in Frontiers of Agricultural Science and Engineering (DOI: 10.15302/J-FASE-2025619).
The core of this technology lies in the integration of spectral data and artificial intelligence (AI). Traditional spectral analysis relies on manual feature extraction, which is inefficient. In contrast, machine learning algorithms can automatically screen key wavebands from massive datasets and maintain high recognition rates even in complex backgrounds. For example, after deploying lightweight convolutional neural network models on UAVs, real-time weed distribution maps can be generated, guiding farmers to complete precise herbicide application within 24 hours. A pilot study in India showed that after adopting this technology, rice field weeding costs decreased by 25% and yields increased by 15%, verifying its practical application value.
Nevertheless, the popularization of hyperspectral imaging technology still faces challenges. Currently, a complete hyperspectral imaging system is relatively expensive, and data processing requires professional knowledge. Additionally, factors such as changes in field lighting and differences in crop growth stages may affect spectral stability. The research team indicates that future efforts should focus on developing more affordable sensors and simplified data analysis tools. Meanwhile, integrating satellite remote sensing and ground-based Internet of Things (IoT) to build an air-space-ground integrated monitoring network will enable small-scale farmers to afford and operate this technology.
With the maturity of technology, hyperspectral imaging can not only be used for weed identification but also for monitoring crop diseases, nutrient status, and yield prediction. It promotes the transformation of agriculture from experience-based cultivation to precision smart agriculture. Against the backdrop of intensifying climate change and growing food demand, such technologies will become crucial supports for safeguarding global food security, ensuring efficient utilization of every mu of rice field, and protecting the sustainable production of food on the plate.
DOI: 10.15302/J-FASE-2025619