استخراج ویژگی های طیفی برای تخمینی از محتوای نیتروژن کل خاک بر اساس الگوریتم بهینه سازی کلونی مورچه اصلاح شده Spectral features extraction for estimation of soil total nitrogen content based on modified ant colony optimization algorithm

1. Introduction Soil is the primary support for soil-grown crops. It is an important medium for plant root extension and the main nutrient source for crop growing. The main soil nutrients include TN, OM, available potassium, and available phosphorus (Chacón Iznaga et al., 2014; Sinfield et al., 2010). Among those soil nutrients, soil nitrogen (TN and available nitrogen) plays the most important role in promoting the growth of leaf, root, and stem and is a decisive factor to the crop yield (Bansod and Thakre, 2014). Excessive nitrogenous fertilization however will cause environmental pollution and crop distortion of growth and quality. The amount of nitrogen fertilizer applied therefore needs to be precisely controlled to ensure the crop yield and environmental protection. The fundament of precision fertilizing is effectively acquiring the soil information in the field. Rapid and precise acquisition of soil nitrogenous information in farmlands hence becomes increasingly important. The conventional method of detecting soil nitrogen content usually takes several days and consumes toxic chemicals. The conventional method also has several disadvantages, such as high requirements for detection personnel, expensive testing equipment, low efficiency, and environmental pollution (Debaene et al., 2014; Florinsky et al., 2002; Kuang and Mouazen, 2013; Moore et al., 1993; Nocita et al., 2014). By contrast, spectral analysis techniques are based on the internal relations between radiation energy and the composition and structure of matters. According to the characteristic spectra of matter, the target concentration or properties can be determined rapidly without chemicals (Igne et al., 2010; Vohland et al., 2011). For soil, the spectral information related to most of the organic radical groups containing hydrogen was in the NIR region (Li, 2006). NIR spectroscopy is a rapid, non-destructive, and non-pollutant testing method that plays an growing important role in soil nutrition measurement and exhibits extraordinary development potential in applications of soil TN content detection (Chang et al., 2001; Lucà et al., 2017; Morellos et al., 2016). In the detection of soil TN content with NIR spectroscopy, the spectra of soil samples are first measured, and the NIR spectral data are then used as the input variables to establish the prediction models. Modern spectrometers possess high spectral resolution, and spectral data measurement generally involves hundreds or thousands of wavelength variables. Three kinds of information variables are involved in measurement of such superlarge-scale data. One is the effective informative variable, which can improve the model predictive ability because it reflects the characteristics of the target substance in the NIR region. The second is redundant or interfering variable, which is related with other targets. The last one is uninformative variable, which is irrelevant to the target material and usually caused by the measurement environment, such as noise. If the prediction model was established by the entire-spectrum information, then the latter two kinds of variables would increase the computation complexity and reduce the target prediction accuracy of the model.