基于数码相片和颜色空间转换的滨海土壤盐渍化定量估算

土壤盐渍化是土壤退化的重要原因之一，快速精确地监测土壤盐渍化对农业可持续发展和生态环境保护有积极作用。提出一种基于数码相片的滨海地区表层土壤盐分的定量估算方法，旨在复杂天气状况下快捷方便的获取土壤盐分信息。以江苏盐城沿海地区裸露地表土壤作为研究对象，在晴天和多云天气下全天时采样和拍照获取52个土壤样本和相片。土壤样品通过室内测试获取土壤电导率（EC），pH值和土壤含水量等参数。利用RStudio软件对土壤相片进行处理，首先从相片中提取RGB三种颜色参数，再通过颜色空间转换关系计算另外5种颜色空间（HIS，CIEXYZ，CIELAB，CIELUV和CIELCH），每个颜色空间有三个颜色参数，加上RGB颜色空间共有18个颜色参数，其中CIELAB，CIELUV和CIELCH中L参量表示的意义和数值相同，因此6个颜色空间共有16个颜色参数。土壤电导率与颜色参数相关分析结果表明，相片颜色的纯度和亮度与土壤电导率之间的相关系数较高，并达到了极显著水平，相片颜色的色相与土壤电导率之间的相关性较低，且未达到显著水平。随机抽取70%的样本数据并用随机森林方法对土壤盐分含量进行建模，采用留一法（LOOCV）进行交叉验证，再用余下30%的样本数据进行精度检验，重复100次以获取精度最高的模型。最终获取估算土壤盐分的随机森林模型，验证集数据的模型精度达到R2_val=0.75，RMSE_val=3.52，RPD_val=2.02。对颜色参数进行重要性分析发现，颜色纯度对模型的重要性最大，其次是颜色亮度，色相的贡献较小。综上，利用数码相机获取表层土壤相片，通过颜色空间转换得到的颜色参数为有效估算滨海土壤盐分含量提供一个新思路。该研究对近地表参数定量估算提供了新视角，将来结合无人机平台能够为精准农业和滨海生态环境的精准管理提供技术支持和有效手段。

Coastal Soil Salinity Estimation Based Digital Images and Color Space Conversion

Soil salinization is one of the most important reasons for soil degradation. Rapid and accurate monitoring of soil salinity has positive effects on sustainable agricultural development and ecological environment protection. This study proposed a new method of surface soil salinity estimation in coastal areas based on digital photographs to obtain soil salinity information quickly and conveniently under complicated weather conditions. 52 bare surface soil samples and photographs were collected under sunny and cloudy weather in the coastal area of Yancheng, Jiangsu province. Parameters such as soil electrical conductivity (EC), pH value and soil water content were measured in the lab. Using RStudio software for photo processing, firstly, three color components were extracted from RGB color space, then five color spaces (HIS, CIEXYZ, CIELAB, CIELUV, and CIELCH) were obtained from color space conversion. Three parameters were extracted from each color space. Hence there were 16 parameters from 6 color spaces for CIELAB, CIELUV, and CIELCH having the same parameter L. The correlation analysis of soil EC and color parameters indicated that the color purity and brightness were significantly correlated with soil EC, while color hue was insignificantly correlated with soil EC. Random forest and leave one out cross validation methods were used to establish soil EC estimation model with randomly 70% dataset, and the rest 30% dataset was used for validating. Repeated 100 times to get the optimal model, and finally, the accuracy of the best model reached R2_val=0.75, RMSE_val=3.52, RPD_val=2.02. By analyzing the importance of color parameters, we found that color purity and color brightness contributed most to the model, and color hue contributed relatively little. To sum up, the color parameters obtained from digital images provided a new approach for soil salinity estimation effectively. Combined with the unmanned aerial vehicle, this study proposed a new perspective for quantitative assessment of surface parameters, which would provide technical support and effective means for the precise management of precision agriculture and coastal ecological environment in future.