Residual Densely Connected Progressive Attention Based Transnet for Crop Recommendation Using Soil Conditions

Agriculture is the major contributor to a country's economy and the employment sector. Selecting the wrong crops that are not appropriate for given soil and environmental circumstances is one of the main challenges for farmers, causing low productivity and soil fertility problems over a long period. Building on the fusion of cutting-edge machine learning approaches and metaheuristic optimization algorithms, this study presents a fresh crop recommendation model meant to solve such major problems. The model starts with gathering input data from public datasets, including significant agricultural factors like nitrogen (N), phosphorus (P), potassium (K), pH value, temperature, humidity, and rain circumstances dramatically impacting crop suitability. In the preprocessing stage, z-score normalization is used to improve consistency and model accuracy; input data will be normalized, and unwanted data will be removed. The Dung Sparrow Optimization (DSO) algorithm, which simulates adaptive dung beetle and sparrow behaviors, is used for feature selection to maximize exploration and exploitation. The procedure improves the choice of the most insightful features for improving learning speed, performance, and convergence rate. Using Residual Densely Connected Progressive Attention Based Transnet (RDPAT), an advanced transformer based deep learning approach developed for precise crop prediction and soil nutrient analysis, the selected features are processed. The RDPAT system employs residual connections, dense layers, and progressive attention mechanisms to extract and rank informative patterns. Experimental results show the high performance of the model, with 99.26% accuracy, 96.25% recall, 98.93% precision, and 18.76s of execution time, as well as its efficacy and reliability for practical agricultural use.