Skip to content

AI / Machine Learning · Dissertation

Early Disease Detection in Broiler Chickens

An AI-powered system for early detection of disease in broiler chickens using audio and visual analysis.

  • Python
  • Machine Learning
  • Computer Vision
  • Audio Classification
  • Raspberry Pi
  • Firebase
  • Android
Type
AI / Machine Learning · Dissertation
My role
Architecture, development & delivery
Core stack
Python, Machine Learning, Computer Vision
Audio and image analysis results from the broiler disease detection system

Context

The Problem

Disease can spread quickly through a broiler flock. By the time symptoms are obvious to the eye, losses can already be significant, especially for small-scale farmers without regular veterinary support.

Early signs often show up first in changes to the birds' sounds and droppings, but monitoring these continuously by hand is not practical.

What I built

The Solution

I designed and built a system that monitors the flock continuously with a microphone and camera, analyses the data with machine learning models running on a Raspberry Pi, and sends results through Firebase to an Android app so the farmer can act early.

Capabilities

Key Features

My Computer Science final-year dissertation project: an end-to-end system that combines machine learning on audio and images, an edge device on the farm and an Android application that alerts the farmer.

  • Audio-based detection

    Analyses chicken vocalisations to flag sounds associated with respiratory illness.

  • Visual droppings analysis

    Classifies images of droppings for visual signs of disease.

  • Machine learning models

    Trained models for both the audio and the image pipeline.

  • Edge deployment

    Inference runs on a Raspberry Pi close to the flock.

  • Firebase communication

    Results are synced to the cloud in real time.

  • Android application

    Farmers receive alerts and see flock status on their phone.

Deep dive

Machine Learning Approach

The system uses two complementary models: one classifies audio recorded in the poultry house, the other classifies images of droppings. Combining sound and visual signals gives the farmer two independent early-warning indicators.
  • Audio pipeline: recordings are pre-processed into features and classified as healthy or as indicating possible illness
  • Vision pipeline: droppings images are classified for visual signs associated with disease
  • [Add model architecture, dataset size and preprocessing details]

Deep dive

Hardware

The models run on a Raspberry Pi connected to a camera and a microphone, so analysis happens on the farm and only results need to be sent to the cloud.
  • Raspberry Pi (edge inference)
  • Camera for droppings images
  • Microphone for flock audio

User experience

Interface & Screenshots

Select any screen to view it full size.
  • Audio-based detection
  • Android app alerts
  • Visual droppings analysis

1 / 3

Audio waveform with classification output

Audio-based detection

Engineering

System Architecture

  1. 01Sensing

    • CameraDroppings images
    • MicrophoneFlock audio

  2. 02Machine Learning

    • Audio modelSound classification
    • Vision modelImage classification

  3. 03Edge Device

    • Raspberry PiOn-farm inference

  4. 04Cloud

    • FirebaseReal-time results sync

  5. 05Farmer

    • Android appAlerts & flock status
Data is captured on the farm, analysed locally on a Raspberry Pi, and only the results are pushed through Firebase to the farmer's Android app.

Tools

Technology Stack

Machine Learning
PythonAudio classificationImage classification
Edge hardware
Raspberry PiCameraMicrophone
Cloud & Mobile
FirebaseAndroid (Java)

Engineering decisions

Challenges & How I Solved Them

  1. 01Running ML on constrained hardware

    The challenge

    A Raspberry Pi has limited memory and compute compared to the machine used for training.

    How I solved it

    Kept inference on the edge lightweight and sent only results to the cloud instead of raw audio and images.

  2. 02Noisy real-world audio

    The challenge

    Poultry houses are noisy environments, which makes audio classification harder than in clean recordings.

    How I solved it

    Pre-processed recordings before classification so the model focuses on the relevant parts of the signal.

  3. 03Getting results to the farmer

    The challenge

    A detection is only useful if the farmer sees it quickly, wherever they are.

    How I solved it

    Used Firebase as a real-time bridge between the edge device and an Android app.

Ownership

My Contribution

Designed and built the full system as my final-year dissertation: data pipeline, machine learning models, Raspberry Pi deployment, Firebase integration and the Android application.

  • Researched the problem and defined the system's scope
  • Prepared data and trained the audio and image models
  • Deployed inference on a Raspberry Pi
  • Integrated Firebase for real-time communication
  • Built the Android application for farmers
  • Documented and presented the work as my dissertation

Outcomes

Results

Only verified outcomes are listed here.
  • [Add model evaluation results, e.g. accuracy on a held-out test set]
  • [Add dissertation grade or recognition if applicable]

Want to know more about this project?

I'm happy to walk you through how it works and the decisions behind it.

Get in touch