AN INTELLIGENT DEEP LEARNING-BASED MODEL FOR WEED DETECTION AND CLASSIFICATION USING THE YOLOV8N ARCHITECTURE

The weed invasion is one of the most severe challenges to sustainable rice production, which diminishes crop production and over depends on herbicides, which usually result in environmental degradation and appearance of herbicides-resistant species of weeds. Precision agriculture requires early and accurate detection of weeds, which have limitations on their accuracy of detection, false alarms in cases of background occlusion, and differences in the appearance of the weed across farm settings. This paper will suggest a smart deep learning model that detects and classifies weeds based on the YOLOv8n architecture created within the Extreme Programming (XP) approach. The testbed was a case study rice farm in Ndibinagu-Uzam, Ihuokpara, Enugu State, Nigeria. A total of 10,000 annotated images of the most common species was collected as rice weed, specifically barnyard grass, rice flatsedge, and blistering ammannia in different lighting and field conditions. The labels were given to the images with the help of Roboflow and were organized in a SQLite database to train and validate the model. Using this data, the YOLOv8n transfer learning model was trained, and its performance was measured according to such standard metrics as accuracy, precision, recall, and F1-score. The experimental findings have shown that the proposed model has achieved a total accuracy of 94.6%, precision of 92.8%, recall of 93.5% and F1-score of 93.1% indicating that the proposed model is robust when detecting small objects and minimizing false alarms due to overlapping and occlusion. The results indicate that the system can effectively be used as a dependable decision support tool in real time monitoring of weeds, lessening herbicide addiction, and enhancing rice harvest in small-scale farms. The paper concludes that a solution of deep learning with locally gathered agricultural data is context-specific to sustainable management of weeds, and further efforts will involve real-time Simple Mail Transfer Protocol(SMTP) and aerial implementation to implement this solution in large scale.