Artificial Intelligence Driven Unmanned Aerial Vehicle Based Hyperspectral Imaging for Agriculture

The increased demand for food production and the speed of climatic changes in recent years pose challenges to the agriculture sector. Plant phenotyping is one of the crucial areas of crop improvement programs, where breeding scientists are working to develop new crop varieties that are climate-resilient, stress-tolerant, and produce high yields with fewer inputs. However, the major bottleneck in phenotyping is the assessment of thousands of genotype varieties under field conditions. Traditional phenotyping methodologies are expensive, time-consuming, labour-intensive, destructive, and require a biological understanding of the monitored data. Non-invasive methods, such as Unmanned Aerial Vehicle (UAV) based imaging techniques, can accelerate phenotyping studies due to their extensive coverages, low revisiting periods, and high spatial resolutions. RGB imaging captures spatial information alone, capable of studying the visual traits only, whereas the limited spectral data acquired by multispectral cameras is inefficient in studying complex traits. However, phenotyping for complex traits such as abiotic stress adaptation, like early water and nutrient stress identification, can be explored more precisely through the rich spectral information captured by Hyperspectral Imaging (HSI) sensors in hundreds of narrow spectral bands. The analysis of HSI data is challenging due to the high dimensionality and spectral redundancy, which can be addressed more effectively by machine learning/deep learning (ML/DL)-based techniques. The main objective of our research is to analyze crop HSI data (in 400-1000nm range) acquired from UAVs using ML/DL-based techniques for the following: