Precision weed detection using yolov11 for enhanced agriculture management

This work provides a weed detection system based on deep learning, utilizing the cutting-edge YOLOv11 object detection algorithm for the classification and identification of several species of weeds with very high accuracy in real-time. This work aims to help farmers automate the detection of weeds in order to increase crop yield, cut down labor costs, and chemical abuse. A tailored dataset of labeled images of 31 different types of weeds was prepared and annotated through bounding box methods, which were further augmented with data augmentation for enhancing the generalizability of the model. The YOLOv11 model was trained with optimal hyperparameters, reporting a mean Average Precision (mAP) of 91.4% at an Intersection over Union (IoU) of 0.5 and an average detection latency of 87 milliseconds per frame, allowing for high-speed capability appropriate for field use. The trained model was incorporated into an intuitive web-based tool developed using Flask, which records real-time input using webcam, identifies the weed species, and returns detailed information such as the weed name, botanical family, morphological features, suggested removal techniques, and suitable herbicide recommendations. The system was tested under diverse lighting and background settings, exhibiting strong performance and stable accuracy. By connecting advanced object detection with realistic agronomic assistance, the proposed system offers an efficient and scalable means for precision weed management to support sustainable agricultural practice and better decision-making by farmers.