Improving remote sensing estimation accuracy of pasture crude protein content by interval analysis

Abstract: Crude protein is the key indication for evaluation of the quality and feeding value of pasture grass. Estimating crude protein content of pasture grass is necessary for monitoring grassland nutrition status, sustainable utilization and management of grassland resources, eventually preventing grassland degradation. Hyperspectral remote sensing technology is supplied as a new approach for scientists to study properties and processes of ecosystems and their inner biochemical content variation. In view of the limitation of ground remote sensing and astronautics remote sensing, we try to construct estimation model of pasture crude protein content based on the hyperspectral aerial airship imaging system, in order to meet the application needs of smart animal husbandry. In view of the uncertainty problems of traditional biochemical parameter inversion models and practical application needs in agriculture and animal husbandry production, we propose a multi-step pasture crude protein content estimation model, which combined the equal width interval division method, stepwise discriminant analysis and Fisher discriminant method. An experiment was designed to determine whether pasture crude protein content could be predicted by means of the developed strategy. Jinyintan grassland, a typical prairie in Haiyan County, Qinghai Province was chosen as the research area. The hyperspectral data were acquired with the hyperspectral mapping system installed on an airship (named ASQ-HAA380), which was developed by our research group. Pasture crude protein samples were collected at the same time and analyzed in Qinghai University. The results show that the proposed model can accurately estimate the crude protein content of pasture. The test accuracy of the 3 models with different interval numbers (3, 5, and 7) for all samples is 95%, 95% and 85% respectively, while their corresponding cross-check accuracy is 90%, 80% and 65% respectively. Compared with the traditional stepwise linear regression method, the estimation accuracy also has a great improvement (overall test accuracy is increased by 18.7%-70%, and cross test accuracy is increased by 20%-62.5%). The selected bands of the 3 models with different interval numbers (3, 5, and 7) are 870, 815, 802, 737, 391 nm; 988, 391, 398, 405, 548 nm; and 870, 815, 946, 888, 839 nm respectively. In addition, we can adjust content interval range according to different application requirements. And our experimental results indicate that the model accuracy is inversely proportional to interval number. In general, this paper has successfully realized the accurate estimation of the crude protein content of pasture with hyperspectral aerial airship imaging data, which provides reference and technical basis for quantitative estimation of crude protein content and efficient implementation of precision livestock husbandry based on hyperspectral images, and also lays the foundation for the development of intelligent livestock husbandry in the future.