Abstract:In order to optimize nutrient management and ensure normal plant growth, UAV remote sensing technology was used to efficiently and non-destructively collect crop seedling information in the field, and to monitor the leaf area index (LAI) and the relative chlorophyll content (SPAD) of oilseed rape during the seedling stage. It is difficult to balance the monitoring efficiency and monitoring accuracy due to the constraints of flight altitude and image resolution of UAVs. A super-resolution reconstruction method was adopted to integrate the high-resolution images taken at lower flight altitudes and reconstruct the images taken at higher flight altitudes, so that LAI and SPAD could be monitored by the flight images taken after the modeling was completed. Three nitrogen fertilizer gradients, three sowing periods, and three planting densities were set up, and the UAV was used to collect the images of oilseed rape seedlings at 20m and 40m flight altitudes respectively in seedling stage, and SRRestnet method was used to analyze the seedling images at 40m and 40m flight altitudes respectively. SRRestnet method, and super-resolution reconstruction was performed on the 40m images. Based on the three combinations of features extracted from the 20m, 40m and 40m reconstructed images, three machine learning methods, namely partial least squares regression (PLSR), random forest (RF), and support vector regression (SVR), were utilized to monitor LAI and SPAD. The results showed that the super-resolution reconstructed images performed well in phenological seedling monitoring, and PLSR monitoring of LAI and RF monitoring of SPAD had the highest monitoring accuracy, and the operational efficiency of the 40m reconstructed images was 48.6% higher compared with that of the 20m images.
