Literature Review

Introduction to Literature Review

The purpose of this literature review is to provide a comprehensive overview of the existing research and technological advancements in the field of AI-based animal detection and counting. By examining previous studies and developments, this review not only contextualizes the project within existing literature but also identifies gaps, such as the limited focus on rabbit detection, and demonstrates how this project bridges those gaps through innovative applications of deep learning.

Historical Background

The application of AI and computer vision in various domains has been a topic of research for several decades. Early work in this field focused on developing basic image processing techniques and pattern recognition algorithms. For instance, the development of convolutional neural networks (CNNs) by LeCun et al. in the late 1980s marked a significant milestone in the field of computer vision. These early CNN models laid the groundwork for more sophisticated image recognition systems that followed.

Recent Advances in AI and Computer Vision

In recent years, significant advancements have been made in AI and computer vision, driven by the advent of deep learning. Key research papers such as Krizhevsky et al.'s "ImageNet Classification with Deep Convolutional Neural Networks" (2012) and He et al.'s "Deep Residual Learning for Image Recognition" (2015) have demonstrated the power of deep neural networks in achieving state-of-the-art performance on various image recognition tasks.

Object detection, a subset of computer vision, has also seen remarkable progress. Techniques such as Region-based CNN (R-CNN), You Only Look Once (YOLO), and Single Shot MultiBox Detector (SSD) have revolutionized the way objects are detected and localized within images. These models have been extensively used in various applications, from autonomous driving to facial recognition.

TensorFlow, a widely-used framework in deep learning, was leveraged in this project due to its flexibility, community support, and optimization features. The YOLO (You Only Look Once) family of object detection algorithms was also considered, but TensorFlow provided a robust ecosystem for training and deploying customized models for rabbit detection in real-time.

AI in Agriculture

By applying AI techniques in agriculture, this project demonstrates the potential of deep learning models to improve productivity in animal husbandry. Using real-time rabbit detection, this project addresses key challenges in population counting and monitoring. Below is an example of real-time detection implemented on rabbit farm footage.

real time rabbit detection

Rabbit Detection and Counting

The developed system achieved robust detection and counting capabilities, overcoming challenges such as rabbit occlusions and varying lighting conditions. Below is a demonstration of model performance after iterative improvements.

Loss curves schedules showing iterative model improvement during training.

learning rate schedules

Gaps and Challenges

Despite the progress in AI and computer vision, several gaps and challenges remain in the context of rabbit detection and counting:

1. Limited Research: There is a paucity of specific research focused on rabbit detection and counting, necessitating the adaptation of techniques from related fields.

2. Real-time Processing: Achieving real-time detection and counting in dynamic farm environments poses computational and algorithmic challenges.

3. Accuracy and Robustness: Ensuring high accuracy and robustness in varying lighting conditions and occlusions requires further refinement of models.

This project addresses critical gaps such as the lack of specific research on rabbit detection and the challenges of real-time monitoring in dynamic farm environments. Iterative improvements, real-world deployment, and rigorous testing have ensured that the developed system is both accurate and robust.