Evaluating Tillage Quality under Varying Speed and Depth Using YOLOv7-Based Image Analysis

Soil tillage is a critical agricultural practice that creates favorable conditions for seedbed preparation and plant growth. This study presents an innovative application of artificial intelligence (AI) in agriculture by employing the YOLOv7 algorithm to classify and assess post-tillage soil surface conditions, a domain underexplored in current research. The integration of mechanical operation parameters with AI-based image classification enables optimization of tillage quality and mitigation of soil compaction, highlighting the novelty of this approach. The study aims to improve the efficiency of moldboard plow operations by examining the effects of tillage speed and depth on soil clod distribution, fuel consumption, and power requirements. The YOLOv7 model was used to analyze soil surface imagery and identify optimal tillage outcomes characterized by minimal soil clod presence. Results indicate that increasing tillage speed and depth led to higher fuel and power consumption. Conversely, smaller soil clod sizes, indicative of better surface quality, were achieved at higher speeds and shallower depths. Model performance demonstrated strong accuracy: recall of 78.5%, precision of 82.0%, and F1-score of 80.2%. The mean Average Precision at 0.5 IoU (mAP@0.5) reached 77.4%, with validation and test set values of 78% and 75%, respectively. These results confirm the effectiveness of YOLOv7 in detecting and localizing soil clods, enabling real-time assessment of tillage quality. The proposed framework supports data-driven decision-making in precision agriculture, enhances operational efficiency, and promotes improved seedbed conditions for optimal crop establishment.