<p>Soil is a crucial component in agricultural productivity and environmental health, as various agricultural functions, such as nutrient cycling, water regulation, and carbon sequestration. The Estimation of soil properties is vital for optimizing crop production in sustainable agriculture. However, classical techniques for soil analysis often face significant difficulties in terms of high costs, labor-intensive sampling methods, and limited ability to capture spatial and temporal variability. The primary objective is to introduce Skill Aquila Optimization_LeNet (SAO_Lenet) for soil property prediction. Initially, soil data from a specific dataset is obtained and Z-score normalization is applied to normalize the dataset. Subsequently, feature fusion is achieved through a Siamese Convolutional Neural Network (SCNN) combined with Jeffrey Divergence. Data augmentation is conducted to improve dimensionality using the Synthetic Minority Oversampling Technique (SMOTE). The final step involves predicting soil properties into Soil Organic Carbon (SOC), Calcium (Ca), phosphorus, and sand using the Lenet architecture. The Lenet model is trained with the Skill Aquila Optimization Algorithm (SAO), a newly developed approach that integrates the Skill Optimization Algorithm (SOA) and the Aquila Optimizer (AO). The designed SAO_Lenet has shown remarkable results in soil property prediction, attaining an accuracy of 91.889%, a True Positive Rate (TPR) of 90.416%, a True Negative Rate (TNR) of 92.876% and a False Negative Rate (FNR) of 9.584%.</p>