Abstract: Abstract: This study aims to detect the development of dry-shrinkage cracks on the farmland soil surface using Electrical Resistivity Technology (ERT). The Simultaneous Successive Linear Estimation (SimSLE) coupled with the hierarchical clustering was also used to invert the morphology of the dry-shrinkage cracks. The experimental and simulated images were then compared to analyze the crack morphology. It was found that the resistance of the soil around the cracks increased sharply, where there were some cracks in the soil, while the resistance of the soil without cracks still maintained a slowly increasing trend. This phenomenon was set as a reference for the formation of cracks. After that, the estimated resistivity was obtained by inversion using the SimSLE algorithm, further to represent the development state of soil shrinkage cracks. The evaluation of fracture images between the natural cracks and the inversion was made using the geometrical properties of Minkowski numbers Mk for the quantitative analysis, where Mk was used to describe the structure of fissures. Three parameters were utilized for the Minkowski number of fissures on the soil surface: the area A, the length L, and the Euler number E of the fissure, representing the fissure area, fissure length, and connectivity in the two-dimensional fissure structure, respectively. The Minkowski density of the fracture images and the simulated binary networks were quantified by the open-source code realized by MATLAB software. Three indicators were selected to evaluate, including the coefficient of determination (R2), the consistency index (IA), and the Root Mean Square Error (RMSE). The results show that the simulated images presented accurate to capture the shape and position of the fracture. The area, length, and Euler number density IA of the test and simulation images were all greater than 0.9, and the R2 values were greater than 0.6, but the RMSE values were small. The morphological characteristics of dry-shrinkage cracks in farmland soil were more accurately to continuously update the partition mean and boundary after coupling hierarchical clustering. The simulation images were better after clustering and partitioning. The partition of hierarchical clustering from the resistivity was estimated to obtain via the inversion and calculation of the uniform resistivity as the initial input values in the traditional SimSLE algorithm. The Euclidean distance criterion was used to obtain the optimal number of clusters for the dataset of estimated values. Then, the mean of the estimates for each region was calculated to adjust the covariance matrix. Finally, the updated mean and covariance were used as the prior information for the next round of iterative calculation. As such, the computational accuracy was improved to identify the small-scale spatial variability within soil layers. After the introduction of clustering partition, the R2 and IA of Minkowski density increased by 9.76%-18.5%, and 0.93%-1.47%, respectively, whereas, the RMSE was reduced by 12.1%-21.1%, indicating an improved accuracy of the prediction. Therefore, effective detection of soil fissures can greatly contribute to preventing the occurrence of cracks from the damaged integrity of the soil structure, even various engineering geological disasters. The finding can provide an ideal way to characterize the fractures morphologies using non-invasive detection.