|Xu, Tian-Shuang -|
|Xu, Tao -|
|Wu, Wenfu -|
|Zhu, Hang -|
Submitted to: Spectroscopy and Spectral Analysis
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: October 12, 2012
Publication Date: October 20, 2012
Citation: Xu, T., Xu, T., Lan, Y., Wu, W., Zhang, H., Zhu, H. 2012. A method for fast selecting feature wavelengths from the spectral information of crop nitrogen. Spectroscopy and Spectral Analysis. 32:1-5. Interpretive Summary: A wide variety of remote sensing data is available for site-specific management in crop production and protection. However, new approaches are still needed to extract effective spectral information from raw data from sensors, such as hyperspectral remote sensing data. Remotely-sensed data gathered with a handheld spectroradiometer were used for detection of nitrogen status of crops. Through analysis of the remote sensing data with a proposed new method, the number of selected wavelengths for detecting crop nitrogen status was reduced dramatically and the computation was faster and simpler than using the original dataset. These new methods will provide a rapid way for farmers and researchers to assess nitrogen levels in various crops and make fertilizing decisions, which will reduce input cost and the over application of nitrogen to fields.
Technical Abstract: Research on a method for fast selecting feature wavelengths from the nitrogen spectral information is necessary, which can determine the nitrogen content of crops. Based on the uniformity of uniform design, this paper proposed an improved particle swarm optimization (PSO) method. The method can choose the initial particle swarm uniformly and describe the optimization space well by fewer sample points, which is helpful to avoid the local optimum and accelerate the convergence to a meaningful assess of nitrogen content in plants. Then, the method was applied to fast select the nitrogen spectral wavelengths of soybean, cotton and maize. Calibration models based on the partial least square (PLS) method and selected wavelengths were constructed. The results illustrate that compared with the original wavelengths, the number of selected wavelengths decreased about 93 percent, which means the computation is simplified. Also, the precision of PLS prediction mode based on the selected wavelengths improves 34 percent at least, and the prediction ability of calibration model increases greatly. Therefore, the proposed method is both correct and effective and will improve the speed of assessing nitrogen levels in plants.