🐟 CatchGrade — Skipjack Tuna Defect Classification App

An Android mobile application that uses on-device machine learning to classify skipjack tuna quality in real-time, helping fishery inspectors and processors identify defects instantly from a camera photo or gallery image.

---

📋 Table of Contents

• Overview
• Features
• Classification Categories
• Tech Stack
• Architecture
• Screenshots
• Getting Started
• Project Structure
• How It Works
• Contributing
• License
• Author

---

Overview

CatchGrade was developed as a capstone research project to address quality control challenges at the General Santos City Fish Port Complex — the largest tuna landing port in the Philippines. Manual visual inspection of skipjack tuna is time-consuming and prone to human error. CatchGrade automates this process using a trained TensorFlow Lite image classification model deployed directly on-device for instant, offline-capable results.

Problem Statement

Traditional tuna quality assessment relies on manual visual inspection by trained personnel, which is:

• Slow — inspectors can only process a limited number of fish per hour
• Subjective — results vary between inspectors
• Error-prone — fatigue leads to missed defects

Solution

CatchGrade provides a mobile, camera-based classification tool that delivers consistent, instant results — reducing inspection time and improving accuracy for fishery quality control operations.

---

Features

• 📸 Real-Time Camera Capture — Use CameraX back-camera preview to photograph tuna specimens directly
• 🖼️ Gallery Upload — Import existing images from device gallery for classification
• 🧠 On-Device ML Inference — TensorFlow Lite model runs entirely on-device (no internet required)
• 📊 Instant Classification — Results displayed in under 1 second with confidence scoring
• 📜 Scan History — Horizontal scrollable history reel showing all previous scans with thumbnails
• 🔄 Re-Review Scans — Tap any history thumbnail to re-display the image and its classification result
• 📱 Optimized for Field Use — Designed for use in processing plants and port facilities

---

Classification Categories

Category	Description
✅ Healthy	Tuna specimen shows no visible defects — passes quality inspection
⚠️ Skin Discoloration	Abnormal coloring on the skin surface indicating potential quality degradation
🔴 Eye Deterioration	Visible deterioration of the eye area, a key freshness indicator

The TFLite model accepts 224×224 RGB images and outputs probability scores across all three classes.

---

Tech Stack

Layer	Technology
Language	Kotlin 2.0
Platform	Android (minSdk 24 / Android 7.0+)
ML Framework	TensorFlow Lite + TFLite Support Library
Camera	CameraX 1.3.1 (camera-core, camera2, lifecycle, view)
UI	XML Views, ConstraintLayout, Material Design 3
Build	Gradle 8.9.2 (Kotlin DSL), AGP 8.9.2
IDE	Android Studio

---

Architecture

┌─────────────────────────────────────────────┐
│                 MainActivity                │
│  ┌──────────┐  ┌──────────┐  ┌───────────┐ │
│  │ CameraX  │  │ Gallery  │  │  History   │ │
│  │ Preview  │  │ Picker   │  │  Adapter   │ │
│  └────┬─────┘  └────┬─────┘  └───────────┘ │
│       │              │                       │
│       ▼              ▼                       │
│  ┌─────────────────────────┐                │
│  │    Bitmap Processing    │                │
│  │  (resize to 224×224)    │                │
│  └───────────┬─────────────┘                │
│              ▼                               │
│  ┌─────────────────────────┐                │
│  │    TunaClassifier       │                │
│  │  (TFLite Interpreter)   │                │
│  │  tuna_model.tflite      │                │
│  └───────────┬─────────────┘                │
│              ▼                               │
│  ┌─────────────────────────┐                │
│  │  Classification Result  │                │
│  │  + Confidence Score     │                │
│  └─────────────────────────┘                │
└─────────────────────────────────────────────┘

---

Screenshots

📌 Screenshots coming soon — run the app in Android Studio to preview the UI.

---

Getting Started

Prerequisites

• Android Studio Arctic Fox or later (recommended: latest stable)
• JDK 11 or higher
• Android SDK with API Level 24+ installed
• A physical Android device or emulator (camera features work best on a real device)

Installation

1. Clone the repository

   git clone https://github.com/YOUR_USERNAME/CatchGrade.git
   cd CatchGrade

2. Open in Android Studio

   • File → Open → select the cloned CatchGrade directory
   • Wait for Gradle sync to complete

3. Build and Run

   • Connect an Android device (USB debugging enabled) or start an emulator
   • Click Run ▶ or press Shift + F10
   • Grant camera permission when prompted

Testing with Sample Images

If you don't have access to tuna specimens, you can test by:

1. Tap "Upload Picture"
2. Select any fish image from your gallery
3. The model will classify it based on visual features

---

Project Structure

CatchGrade/
├── app/
│   ├── build.gradle.kts              # App-level build config
│   ├── src/
│   │   ├── main/
│   │   │   ├── AndroidManifest.xml   # Permissions & activity config
│   │   │   ├── assets/
│   │   │   │   └── tuna_model.tflite # Trained ML model (2.5 MB)
│   │   │   ├── java/com/example/catchgrade/
│   │   │   │   ├── MainActivity.kt   # Camera, gallery, UI logic
│   │   │   │   ├── TunaClassifier.kt # TFLite inference engine
│   │   │   │   └── HistoryAdapter.kt # Scan history RecyclerView
│   │   │   └── res/
│   │   │       ├── layout/
│   │   │       │   ├── activity_main.xml  # Main UI layout
│   │   │       │   └── item_history.xml   # History thumbnail item
│   │   │       ├── values/
│   │   │       │   ├── colors.xml
│   │   │       │   ├── strings.xml
│   │   │       │   └── themes.xml
│   │   │       └── mipmap-*/         # App launcher icons
│   │   └── test/                     # Unit tests
│   └── proguard-rules.pro
├── build.gradle.kts                  # Root build config
├── settings.gradle.kts
├── gradle.properties
├── gradlew / gradlew.bat            # Gradle wrapper
├── LICENSE
└── README.md

---

How It Works

1. Image Capture

The app uses CameraX to provide a live back-camera preview. When the user taps "Take Picture," an in-memory image is captured and auto-rotated to correct orientation.

2. Preprocessing

The captured or uploaded bitmap is:

• Decoded safely using Software rendering (avoiding Hardware Bitmap crashes)
• Resized to 224×224 pixels (the model's expected input size)
• Normalized from 0–255 to 0.0–1.0 float values per RGB channel

3. Inference

The TunaClassifier loads tuna_model.tflite from the app's assets directory and runs inference using the TensorFlow Lite Interpreter. The model outputs a probability array across 3 classes.

4. Result Display

The class with the highest probability is displayed on a blue banner at the top of the screen. The image and result are also added to the scan history reel for later review.

---

Model Details

Property	Value
Format	TensorFlow Lite (.tflite)
Input	224×224×3 (RGB float32, normalized 0–1)
Output	1×3 probability array
Classes	Skin Discoloration, Healthy, Eye Deterioration
Size	~2.5 MB
Training	Transfer learning (trained offline, exported to TFLite)

---

Contributing

Contributions are welcome! If you'd like to improve CatchGrade:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/your-feature)
3. Commit your changes (git commit -m 'feat: add your feature')
4. Push to the branch (git push origin feature/your-feature)
5. Open a Pull Request

---

License

This project is licensed under the MIT License — see the LICENSE file for details.

---

Author

Dayle Angelo Peña

• 📧 daylepena@gmail.com
• 🎓 B.S. Computer Science — Notre Dame of Dadiangas University
• 📍 General Santos City, Philippines

---

Acknowledgments

• General Santos City Fish Port Complex — domain expertise and research support
• Notre Dame of Dadiangas University — academic guidance
• TensorFlow — open-source ML framework
• CameraX — modern Android camera API