Deep Learning Approach For Weed Detection To Determine Soil Condition

Abstract

Mobile technology has transformed information gathering and accessibility, enabling real-time data collection and analysis through portable devices. Identifying soil conditions is critical in agriculture, yet traditional methods often require specialized tools and can be challenging. The presence of weeds can serve as an initial indicator of soil conditions. This study addresses these challenges by developing a mobile application that utilizes Artificial Intelligence (AI) to determine soil conditions based on weed identification. The application employs YOLOv11, a state-of-the-art object detection model known for its speed, accuracy, and efficiency. YOLOv11's backbone incorporates C3K2 blocks, an evolved CSP bottleneck for efficient feature extraction, while its neck integrates the Spatial Pyramid Pooling Fast (SPFF) module and C2PSA block to enhance spatial attention and focus on critical regions. Data collection and processing were conducted using Google Colab and Roboflow, achieving outstanding performance metrics, including high precision, recall, F1 score, and an impressive 87% mean average precision (mAP). In a comparative analysis of YOLOv11, YOLOv10, and SSD, YOLOv11 stands out with an impressive 87% mean average precision (mAP), demonstrating its superior accuracy in identifying weeds and determining soil conditions. The integration of this model into Android devices allows users to capture weed images and give the soil condition based on the weeds. This application addresses this gap by providing farmers with an efficient and hassle-free solution using artificial intelligence to determine soil conditions based on weeds.