Estimation of nitrogen content in walnut canopy by UAV hyperspectral camera

Walnut is an important nut fruit tree and woody oil tree species in China. With its unique flavor and rich nutritional value, walnut ranks first among the four dried fruits in the world. Fruit expansion stage is the first stage of the development of walnut fruit, such as insufficient nutrition in this stage will directly affect the quality and yield of the later fruit. Therefore, the monitoring and diagnosis of nitrogen content of walnut fruit in the expanding stage is of great significance for controlling tree growth and adjusting fine management plan in time.

In this study, a 400-1000nm hyperspectral camera was applied, and FS60, a product of Hangzhou Color Spectrum Technology Co., LTD., could be used for related research. The spectral range is 400-1000nm, the wavelength resolution is better than 2.5nm, and up to 1200 spectral channels can be reached. The acquisition speed can reach 128FPS in the full spectrum, and the maximum after band selection is 3300Hz (support multi-region band selection).

一、Preliminary preparation

In order to estimate the nitrogen content of walnut canopy by UAV hyperspectral camera, data collection is needed first. Select the appropriate UAV platform equipped with hyperspectral camera, and carry out flight operations in accordance with the predetermined route and height over the Walnut Garden. During the flight, the hyperspectral camera imgrams the walnut canopy at a certain time interval or space interval to obtain a large amount of hyperspectral image data. At the same time, in order to ensure the accuracy and reliability of the data, it is also necessary to collect some reference data simultaneously on the ground, such as the nitrogen content of walnut leaves and canopy structure parameters determined by traditional methods.

二、Results and analysis

Canopy range determination, canopy spectrum extraction and accuracy verification As shown in Figure 2, walnut, soil and shadow overlap to a certain extent in the whole band range of the 5-year old walnut forest remote sensing image. In the band range of 520~600nm, the spectral reflectance of the shadows is less than 0.10: the difference of spectral reflectance of walnut and soil is obviously not overlapping, and the spectral reflectance of both is greater than 0.10 in this range. In the range of 750-1000nm, the spectral reflectance of walnut, soil and shadow is significantly different. The spectral reflectance of walnut is greater than 0.7 in the range of 740-900nm, and the spectral reflectance of other non-target vegetation is less than 0.7. Since the spectral reflectance of walnut can be distinguished from other non-target vegetation in green light and near infrared band, but not in one or some bands, it cannot be calculated in ENVI5.3 software. Therefore, in order to facilitate the smooth extraction process of walnut canopy range, the maximum spectral reflectance of walnut canopy in green light and near infrared band is selected in this study Bw(550.7) and B(779.4) were classified and identified to determine the canopy range. Walnut tree, soil and shadow are defined in ENVI5.3 software, that is, when the spectral reflectance at B(550.7) is less than or equal to 0.10 and the spectral reflectance at B(779.4) is less than or equal to 0.20, the shadow is identified and eliminated. When the spectral reflectance at B(550.7) is greater than 0.10 and B; When the spectral reflectance at (779.4) is less than or equal to 0.70, it is identified as soil and removed; When the spectral reflectance at B(550.7) is greater than that at.0.10, the spectral reflectance at B(779.4) is greater than 0.70, walnut tree is identified as the target vegetation.

In addition, a support vector machine with good generalization and classification accuracy was used to extract the canopy range, and the accuracy of the canopy range extraction based on spectral features was compared. First of all, in ENVI5.3 software, the ground objects in remote sensing images are divided into walnut tree and other two types (Figure 4), in which the red area is the walnut canopy, and the green area is the other. The separability between the two types of samples was 1.998, and then SVM classifier was selected for supervised classification to obtain the original classification results (FIG. 5a). However, there were often some small patches in the classification results, and its accuracy was difficult to achieve the purpose of final application. Therefore, the Majority small patch processing method was adopted to process the preliminary classification results, and the classification results meeting the actual requirements were obtained (Figure 5b). The accuracy of the classification results was verified, and the Kappa coefficient was 0.997, and the mapping accuracy of the target vegetation walnut was 99.65%. Finally, Matab2014b software was used to overlap the canopy range determined based on spectral features in this study with the canopy range pixels identified by support vector machine method. There were 4257 overlapping pixels in the canopy range, and the number of canopy range pixels selected based on spectral features accounted for 96.77% of the number of pixels in the support vector machine, with a mapping accuracy of 96.43 %, high precision, overlapping results are shown in Figure 6

At present, the application of UAV hyperspectral camera in estimating nitrogen content of walnut canopy is still in the stage of continuous development and improvement. In the future, with the continuous progress of technology, the performance of hyperspectral cameras will be further improved, the spectral resolution and imaging quality will be higher, and the data processing and analysis methods will be more intelligent and automated. At the same time, the development of multi-source data fusion technology, such as the combination of hyperspectral data with lidar data and thermal infrared data, will be able to obtain more comprehensive and accurate growth information of walnut trees, and further improve the accuracy and reliability of nitrogen content estimation. In addition, with the in-depth promotion of the concept of precision agriculture, UAV hyperspectral camera technology is expected to be more widely used in the field of walnut planting, providing strong technical support for the sustainable development of the walnut industry.

In summary, UAV hyperspectral camera, as an advanced remote sensing monitoring technology, has a broad prospect and great potential in the application of walnut canopy nitrogen content estimation. Accurate and rapid estimation of walnut canopy nitrogen content can provide scientific basis for walnut growers to make fertilization decisions, achieve accurate fertilization, improve fertilizer utilization, reduce resource waste and environmental pollution, and promote the high-quality development of walnut industry.