In the heart of Florida, a team of researchers led by Zhengkun Li from the Bio-Sensing, Automation, and Intelligence Laboratory at the University of Florida is revolutionizing peanut farming with a cutting-edge robotic system. This innovation promises to transform the labor-intensive process of yield estimation, offering a glimpse into the future of high-throughput phenotyping in agriculture.
Peanuts, a staple crop with a farm gate value exceeding $1 billion in the United States, have long relied on conventional yield estimation methods. These methods involve digging, harvesting, transporting, and weighing, which are not only time-consuming but also inefficient for large-scale operations. “The traditional methods are labor-intensive and can be quite inefficient, especially in breeding programs where precise yield estimations are crucial,” Li explains.
The research, published in *Smart Agricultural Technology* (which translates to *智能农业技术* in Chinese), introduces an automated robotic imaging system designed to predict peanut yields in the field after digging and inversion of plots. The system employs advanced image analysis techniques, including the Local Feature Transformer (LoFTR) for image stitching and a customized Real-Time Detection Transformer (RT-DETR) for pod detection.
The RT-DETR model, integrated with partial convolution and refined up-sampling and down-sampling modules, achieved impressive accuracy. “Our customized detector achieved a mean Average Precision (mAP50) of 89.3% and a mAP95 of 55.0%, improving by 3.3% and 5.9% over the original RT-DETR model,” Li notes. This enhanced accuracy, coupled with lighter weights and less computation, makes the system highly efficient and scalable.
The workflow involves stitching sequential images of each peanut plot using LoFTR-based feature matching, which avoids replicated pod counting in overlapped image regions. A sliding window-based method then divides the stitched plot-scale image into smaller patches to improve the accuracy of pod detection. In a case study involving 68 plots across 19 genotypes, the system demonstrated a correlation (R2=0.47) between the yield and predicted pod count, outperforming the structure-from-motion (SfM) method.
The implications of this research are profound. By automating the yield estimation process, the robotic system can significantly reduce the time and labor required for yield determination. This not only improves the efficiency of peanut breeding but also has the potential to revolutionize the agricultural sector as a whole. “Our robotic plot-scale peanut yield estimation workflow shows promise to replace the human measurement process, reducing the time and labor required for yield determination and improving the efficiency of peanut breeding,” Li states.
As the agricultural industry continues to evolve, the integration of advanced technologies like robotic imaging and AI-driven analysis will play a crucial role in enhancing productivity and sustainability. This research by Li and his team is a testament to the transformative power of technology in agriculture, paving the way for a more efficient and data-driven future.