Land evaluation is the process of land performance predictions over time based on land uses and soil features. Traditional methods in determining soil features are proved to be time‐consuming and costly. Therefore, in order to overcome these limitations, a simpler automated method using the image segmentation was developed in this study. The method was designed by integrating dynamic region merging and genetic algorithm. An area index was calculated for each soil profile using the automated method. It was used to present the amount of soil coarse particles and thereupon to determine the rating value of text‐structure. Using the method, the mean intersection over union of above 0.7 was obtained for detecting the coarse particles which confirms its suitability. Data analysis showed that (a) compared to the Storie‐land index (R² = 0.71), the Square root‐land index was more correlated to the harvest index (R² = 0.73), and (b) comparing to manual methods, not only the automated text‐structure had a higher correlation with the harvest index (R² = 0.64) but also it decreased the determination time (>3.75 times). Furthermore, among the models developed by response surface methodology for estimation of soil features, the developed model for estimation of soil lime showed the highest accuracy (R² = 0.89). In conclusion, since the developed method is more accurate, more economic and faster than the usual manual methods, it can be widely used in land suitability evaluation.

中文翻译：

---

基于确定土壤质地-结构和土壤特征的图像处理的土地适宜性评估

土地评估是根据土地用途和土壤特征随时间进行土地绩效预测的过程。事实证明，确定土壤特征的传统方法既费时又费钱。因此，为了克服这些限制，在这项研究中开发了一种使用图像分割的更简单的自动化方法。该方法是结合动态区域合并和遗传算法设计的。使用自动化方法计算每种土壤剖面的面积指数。它用于表示土壤粗颗粒的数量，并由此确定文本结构的等级值。使用该方法，为了检测粗颗粒，获得了大于0.7的联合平均交点，从而证实了其适用性。数据分析表明（a）与Storie-land指数（R ²  = 0.71），平方根指数与收获指数相关性更高（R ²  = 0.73），并且（b）与手动方法相比，不仅自动化文本结构与收获指数具有更高的相关性（R ²  = 0.64），但它也减少了测定时间（> 3.75倍）。此外，在通过响应面方法开发的土壤特征估计模型中，开发的土壤石灰估计模型显示出最高的准确性（R ²  = 0.89）。综上所述，由于所开发的方法比常规的人工方法更准确，更经济，更快捷，因此可以广泛用于土地适宜性评价。