Congliang Zhou, Zelin Qian, Brayden Blanchard, Niranjan Baisakh, Jim Wang and Andre Gama

Sugarcane is one of Louisiana’s most valuable commodities. According to the 2025 Louisiana Summary, it was grown on more than 543,000 acres with a total value of $1.51 billion. LSU AgCenter researchers are developing artificial intelligence (AI) and robotics technologies to improve the efficiency and accuracy of sugarcane breeding and research by automating one of the crop’s most labor-intensive field measurements: stalk counting.

Sugarcane breeding is the foundation for developing varieties with higher yield and greater resistance to disease. In Louisiana, every new variety is tested across many locations and years, and selection decisions depend on accurate field measurements collected from hundreds or thousands of plots. Stalk number is one of the most important traits contributing to final sugar yield. In addition, accurate stalk counts are essential for fungicide trials, fertilizer trials, soil health studies and other field experiments.

Typically, stalk counting is still done by hand. Field crews must walk each plot and manually count stalks, requiring significant labor and time. As trial sizes increase and labor becomes more limited and expensive, collecting consistent, timely stalk population data becomes increasingly difficult.

As shown in Figure 1, counts of the same plots varied considerably among different people. For the 30 plots, each 16 feet long, it took a sugarcane breeder 22 minutes and 45 seconds to complete the counting, which is time-consuming. With thousands of plots to evaluate in a season, a rapid, accurate and consistent stalk counting method is needed.

Recent advances in AI, computer vision and robotics offer a practical solution. AI models can analyze field images and automatically detect and count sugarcane stalks, providing consistent measurements across large numbers of plots. This approach has the potential to reduce labor requirements while improving data accuracy and repeatability.

In July 2025, researchers collected smartphone images of sugarcane stalks from two different breeding trials at the LSU AgCenter Sugar Research Station in St. Gabriel and from a fertilizer trial in St. James Parish. As shown in Figure 2, the different trials represented a range of field conditions, including variation in sugarcane stand density, leaf cover, lodging and background interference.

An iPhone 12 was used to collect video data while walking along sugarcane rows. During recording, the camera was positioned close to the row so individual stalks near the plant base could be clearly observed. For the fertilizer trial data, each video covered a 50-foot counting section. For the breeding trial data, each video covered one 6-foot (first line) and 16-foot (second line) long plot. In total, 127 videos were collected. Importantly, these data were collected around the same time manual stalks counts are typically conducted by researchers. Manual stalk counts in the field and video-based stalk counts in the laboratory served as reference data to evaluate the AI counting results.

Researchers designed and trained different AI models for sugarcane stalk counting. Preliminary smartphone-based results show AI can recognize and count sugarcane stalks with good accuracy and stability. As shown in Figure 3, under different trial conditions, most sugarcane stalks were correctly detected by the AI algorithm. The AI algorithm also maintained good counting performance under challenging conditions, including dense stands, occlusion and complex backgrounds. Across all video datasets, the best AI-based counting method achieved approximately 84% counting accuracy. These preliminary results provide a solid foundation for the further development of an automated robotic system that uses cameras and AI to rapidly count stalks for sugarcane research.

As shown in Figure 4, researchers designed and assembled a remotely controlled robotic system capable of navigating sugarcane research trials. The robotic system has a track width of 72 inches and a height of 135 inches, allowing it to cover a single row within a sugarcane plot at the optimal window of the season where the cane is standing and not lodged. Currently, it can operate in the field for up to two hours of continuous data collection.

Future work will integrate the AI-based stalk counting system into the field robotic platform for practical testing. By improving the speed, consistency and scalability of stalk population measurements, this work will directly support sugarcane variety development for Louisiana by helping breeders make better, data-driven decisions. The robotic platform also has the potential to support a wide range of sugarcane research efforts in Louisiana.

Congliang Zhou is an assistant professor in the LSU AgCenter School of Plant, Environmental and Soil Sciences. Zelin Qian is a Ph.D. student in his lab. Brayden Blanchard is an assistant professor based at the AgCenter Sugar Research Station in St. Gabriel. Niranjan Baisakh is an associate professor, and Jim Wang is a professor, both in the School of Plant, Environmental and Soil Sciences. Andre Gama is an assistant professor in the AgCenter Department of Plant Pathology and Crop Physiology.

This article appears in the summer 2026 edition of Louisiana Agriculture.

Figure 1. Stalk counts measured by three counters across experimental plots. Data provided by Brayden Blanchard

Figure 2. Example images of breeding and fertilizer trials. From left, first line breeding trial, second line breeding trial and fertilizer trial. Photos provided by Zelin Qian

Figure 3. Example images of sugarcane stalk detection result using AI. From left, first line breeding trial, second line breeding trial and fertilizer trial. Photos provided by Zelin Qian

Figure 4. Researchers demonstrate a robotic system at the LSU AgCenter sugarcane field day at the AgCenter Sugar Research Station in 2025. Photo provided by Zelin Qian