Yolov8 Based Animal Intrusion Detection for Crops Protection

Abstract

Wildlife intrusion puts limited regions and agricultural fields at risk, resulting in property destruction and crop damage. In order to overcome this difficulty, a sophisticated System uses deep learning to provide protection and monitoring in real time. The system uses the cutting-edge object identification model YOLOv8 to precisely recognize and categorize animals from real time video feeds using OpenCV. It effectively separates animals from people and other objects by using Convolutional Neural Networks (CNNs), guaranteeing accurate detection. In order to help farm owners make better decisions, each discovered animal is labelled with its matching name and confidence score. The technology automatically sends out email notifications when it detects an animal intrusion, lowering manual surveillance efforts and increasing operational efficiency while enhancing security and facilitating quick response. Because of its flexible and adjustable design, it is a dependable way to protect confined areas and agricultural grounds

Introduction

Problem Context:
Wildlife intrusions into agricultural lands and restricted areas cause significant crop damage and financial loss. Traditional manual surveillance methods are inefficient, slow, and ineffective for real-time prevention.

Existing Solutions:
Past research introduced AI-based animal detection systems such as:

• Deep-AID: Uses hybrid neural networks (CNN + RNN) for accurate detection but lacks instant alerting.

• Road safety systems: Detect animals on highways using posture analysis, but struggle under poor lighting and heavy traffic.

• CNN + ResNet models: Offer high classification accuracy but are computationally intensive and slow for real-time use.

• Hybrid optimization frameworks: Improve efficiency with feature selection but lack adaptability and require domain-specific tuning.

Proposed System:
The new Wildlife Intrusion Detection System uses YOLOv8, a state-of-the-art object detection model, integrated with OpenCV and CNNs, offering:

• High-speed, real-time animal detection in live video feeds.

• Immediate email alerts via SMTP to notify farm owners/security upon intrusion.

• Superior performance compared to older models like YOLOv5, SSD, and Faster R-CNN.

System Architecture:

1. Live Video Capture: Strategically placed cameras stream real-time footage.

2. Object Detection: YOLOv8 identifies animals, humans, and other objects.

3. Alert Mechanism: Instantly sends detection info (animal type, timestamp, confidence) via email.

Technical Highlights:

• Trained on diverse datasets with image augmentation for robustness in various lighting and weather conditions.

• Optimized using techniques like weight pruning, quantization (FP-16), CUDA acceleration, and TensorRT for edge devices.

• Real-time performance: processes 15–20 FPS, with an 8.2ms inference time.

Performance Evaluation:

• Accuracy: 92.1% with high precision (91.8%) and recall (92.5%).

• False Positives/Negatives: Low, though small or distant animals may be missed at times.

• Speed: Outperforms Faster R-CNN (145.8ms) and SSD (16.7ms) with significantly faster processing.

System Requirements:

• Hardware: Multi-core CPU, NVIDIA GPU, high-resolution IP cameras.

• Software: YOLOv8, OpenCV, TensorFlow/PyTorch, Python libraries.

Comparative Benchmarking:

• YOLOv8: mAP@50 = 76.5%, inference time = 8.2ms.

• YOLOv5: Lower accuracy and slower (mAP@50 = 72.3%, 12.5ms).

• Faster R-CNN: Highest accuracy (78.9%) but too slow (145.8ms).

• SSD: Fast but lower accuracy and higher false detection rates.

Conclusion

The System successfully demonstrated real-time wildlife detection with high accuracy, low false detection rates, and fast inference speed. Compared to alternative models, YOLOv8 provided an optimal balance of speed and detection accuracy, making it highly suitable for real-time applications in agricultural and restricted areas. The system consistently detected animal intrusions across different environments, ensuring timely alerts for farm owners and security personnel, however, its limitations include reduced effectiveness in extreme environmental conditions, such as dense vegetation, poor lighting, and heavy fog. The reliance on RGB cameras restricts night-time detection, and the absence of SMS or sound based alerts may limit usability in areas with poor internet connectivity. Future improvements can focus on integrating infrared-based detection for night monitoring, expanding alert mechanisms with SMS and sound alarms, and optimizing the model for edge computing using Tensors for efficient deployment on low power devices. These enhancements will improve adaptability, reliability, and scalability for diverse real-world applications.

References

[1] Sajithra Varun S, G Nagarajan, \"A Smart Framework for Animal Intrusion Recognition and Categorization Using Integrated Deep Neural Networks,\" Soft Computing , pp. 1-12, 2023. [2] Kahlon, Gursimran S, \" A smart surveillance-based system for detecting and alerting authorities about cattle lying on roads in hazardous conditions,\" Multimedia Tools , pp. 1-19, 2023. [3] Vatsavayi, Valli Kumari, and Nagaraju Andavarapu, \"Recognition and Categorization of Wild Animals from Video Streams Using an Enhanced Deep Residual Convolutional Neural Network. \" Multimedia Tools and Applications , pp. 33335-33360, 2023. [4] P. K. Velliangiri S, \"An optimized intrusion detection framework utilizing an effective feature selection strategy for improved accuracy,\" Neural Computing , pp. 7925-7939, 2020. [5] M. F. O. F. Y. Normaisharah Mamat, \"Animal Intrusion Detection in Farming Area using YOLOv5 Approach,\" International Conference on Control, pp. 2642-3901, 2022. [6] M. S. R. C. K. Anuvind P E, \"Development and Implementation of an Animal Intrusion Detection System Using Image and Audio Processing,\" 14th International Conference on Computing Communication, pp. 2473-7674, 2023. [7] T. P. R. S. V. K. Chandralekha E, \"Animal Intrusion Detection System: Protected Crops and Promoted Safety using Machine Learning,\" International Conference on Research Methodologies in Knowledge Management, pp. 979-8-3503-0570-8, 2023. [8] J. S. M. S. S. R. Kothai, \"Animal Intrusion Detection And CropProtection Using Yolo And Sound Devices,\" Intelligent Computing and Control for Engineering and Business Systems, 2023. [9] F. S. e. al., \"Identification of animals and recognition of their actions in wildlife videos using deep learning techniques,\" Ecological Informatics, p. 101215, 2021. [10] R. Chandrakar.,\"Animal Detection and Recognition in Wildlife Sanctuaries Using a Convolutional Neural Network\" International Journal of Recent Technology and Engineering (IJRTE), 2021. [11] S. K. S. Jeevitha, \"A System of Sensor Based Animal Intrusion Recognition System with Image Processing Techniques,\" Third International conference on I-SMAC, 2019. [12] Haricharan N C. B Meenakshi, \"Animal Intrusion Detection and Ranging system using YOLOv4 and LoRa,\" International Conference on Power,, 202. [13] P. K. Panda, \"Implementation of a Wild Animal Detection Model Based on Internet of Things,\" Second International Conference on Artificial Intelligence and Smart Energy (ICAIS), 2022. [14] J. S. K. S. Leela V, \"Animal Intrusion Detection System in Agriculture Using Deep Learning,\" 13th International Conference on System Modeling & Advancement in Research Trends, 2024. [15] K. N. Renuka N, \"Temporal Convolutional Network-Based Animal Intrusion Detection Model for Agriculture \" International Conference on Pervasive Computing and Social Networking (ICPCSN), no. 10.1109/ICPCSN62568.2024.00018, 2024.

Copyright

Copyright © 2025 Dr. Mallikarjuna A, K Akshay Kumar, Kavana K M, L Sametha Bai, Mahesh Ghorpade M. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.