Skip to content
BranchesTags

Repository files navigation

🏛️ Rangkuman Arsitektur Sistem — LeafDiseaseDetector

Aplikasi Android Edge AI untuk deteksi penyakit daun tanaman (Kentang & Tomat) menggunakan model ShuffleNetV2 1.0x yang berjalan 100% offline di perangkat.

1. Informasi Umum Proyek

Item Detail
Nama Aplikasi LeafDiseaseDetector
Package com.randy25.leafdiseasedetector
Bahasa Kotlin
Min SDK 24 (Android 7.0)
Target SDK 36 (Android 15)
Build System Gradle KTS + Version Catalog (libs.versions.toml)
Pola Arsitektur Activity-based (View Layer) + Helper Classes

2. Diagram Arsitektur

2.0 Overview (Garis Besar)

Diagram berikut menunjukkan alur sistem secara keseluruhan dalam 4 blok utama:

graph TB
    USER["Pengguna"]
    UI["UI Layer (Camera & Result Screen)"]
    ENGINE["Inference Engine (Preprocessing + TFLite)"]
    MODEL["On-Device Model (ShuffleNetV2 INT8)"]
    LOG["Monitoring & Logging"]

    USER -->|"Arahkan kamera / Pilih gambar"| UI
    UI -->|"Bitmap gambar"| ENGINE
    ENGINE -->|"Load & Jalankan"| MODEL
    MODEL -->|"Probabilitas 13 kelas"| ENGINE
    ENGINE -->|"Label + Confidence + Latency"| UI
    UI -->|"Tampilkan hasil"| USER
    UI -.->|"Catat metrik"| LOG
Loading

Pengguna berinteraksi melalui UI (kamera real-time atau gallery) → gambar diteruskan ke Inference Engine yang memproses dan menjalankan model TFLite → hasil klasifikasi dikembalikan ke UI → ditampilkan kepada pengguna. Secara paralel, metrik performa dicatat oleh modul monitoring.

2.1 Diagram Detail (Per Komponen)

graph TB
    subgraph "Presentation Layer"
        CA["CameraActivity"]
        RA["ResultActivity"]
    end

    subgraph "ML / Inference Layer"
        ICH["ImageClassifierHelper"]
        BU["BitmapUtils"]
    end

    subgraph "Monitoring & Logging Layer"
        RM["ResourceMonitor"]
        CSV["CSVLogger"]
    end

    subgraph "Assets (On-Device)"
        MODEL["shufflenetv2_int8.tflite<br/>(1.3 MB, INT8 Quantized)"]
        LABELS["labels.txt<br/>(13 kelas)"]
    end

    subgraph "Android Platform APIs"
        CAMERAX["CameraX<br/>(Preview + ImageAnalysis + ImageCapture)"]
        TFLITE["TensorFlow Lite<br/>Interpreter"]
    end

    CA -->|"Real-time frames"| BU
    CA -->|"Capture / Gallery"| BU
    BU -->|"Bitmap 224×224"| ICH
    ICH -->|"Load model"| TFLITE
    TFLITE -->|"Read"| MODEL
    ICH -->|"Read"| LABELS
    ICH -->|"ClassificationResult"| CA
    ICH -->|"ClassificationResult"| RA
    CA -->|"Intent + Extras"| RA
    CA -->|"Log result"| CSV
    CA -->|"CPU / RAM"| RM
    CA -->|"Preview Feed"| CAMERAX
Loading

3. Struktur Direktori Proyek

AndroidAPPProject/
├── app/
│   └── src/main/
│       ├── AndroidManifest.xml
│       ├── assets/
│       │   ├── shufflenetv2_int8.tflite      ← Model TFLite (INT8)
│       │   └── labels.txt                     ← 13 kelas penyakit
│       ├── java/com/randy25/leafdiseasedetector/
│       │   ├── CameraActivity.kt              ← Layar utama + kamera
│       │   ├── ResultActivity.kt              ← Layar hasil klasifikasi
│       │   ├── ImageClassifierHelper.kt       ← Inference engine
│       │   ├── BitmapUtils.kt                 ← Konversi & preprocessing gambar
│       │   ├── CSVLogger.kt                   ← Logging hasil ke CSV
│       │   ├── ResourceMonitor.kt             ← Monitor CPU & RAM
│       │   └── ui/theme/                      ← Jetpack Compose theme (unused)
│       └── res/
│           ├── layout/
│           │   ├── activity_camera.xml         ← Layout kamera real-time
│           │   └── activity_result.xml         ← Layout hasil detail
│           └── values/, drawable/, mipmap-*/
└── build.gradle.kts, settings.gradle.kts

4. Komponen Utama & Tanggung Jawab

4.1 Presentation Layer (Activities)

CameraActivity.kt — Launcher / Layar Utama

Fitur Implementasi
Camera Preview CameraX PreviewView dengan DEFAULT_BACK_CAMERA
Real-time Inference ImageAnalysis (backpressure: KEEP_ONLY_LATEST) → setiap frame diklasifikasi
Capture ImageCapture → klasifikasi static → simpan ke cache → buka ResultActivity
Gallery Picker PickVisualMedia API → decode bitmap → klasifikasi → buka ResultActivity
HUD Overlay Menampilkan: Prediction Label, Confidence %, Latency ms
Threading cameraExecutor (single-thread) untuk capture/analysis, lifecycleScope + runOnUiThread untuk UI

ResultActivity.kt — Layar Hasil Detail

Fitur Implementasi
Tampilan Gambar Membaca captured_image.jpg dari cache
Hasil Klasifikasi Menerima via Intent Extras (label, confidence, latency)
Fallback Jika extras kosong → inisialisasi ulang ImageClassifierHelper → klasifikasi dari cache
Info Ditampilkan Label penyakit, Confidence %, Latency ms, Capture timestamp

4.2 ML / Inference Layer

ImageClassifierHelper.kt — Inference Engine

Komponen inti yang membungkus seluruh pipeline TFLite:

graph LR
    A["Bitmap Input"] --> B["Resize 224×224"]
    B --> C["ImageNet Normalize<br/>mean=[0.485,0.456,0.406]<br/>std=[0.229,0.224,0.225]"]
    C --> D{"Input Type?"}
    D -->|FLOAT32| E1["putFloat(r,g,b)"]
    D -->|INT8| E2["quantizeToInt8()"]
    E1 --> F["TFLite Interpreter.run()"]
    E2 --> F
    F --> G{"Output Type?"}
    G -->|FLOAT32| H1["Read float logits"]
    G -->|INT8| H2["Dequantize logits"]
    H1 --> I["Softmax"]
    H2 --> I
    I --> J["Top-1 Label + Confidence%"]
Loading
Detail Nilai
Model shufflenetv2_int8.tflite (1.3 MB, INT8 Post-Training Quantization)
Input [1, 224, 224, 3] — RGB normalized (ImageNet stats)
Output [1, 13] — 13 kelas (logits → softmax → probabilitas)
Threads 4 CPU threads
Quantization Adaptif: membaca dataType() dan quantizationParams() saat init
Suspend classifyRealtime() dan classifyStatic() berjalan di Dispatchers.Default

BitmapUtils.kt — Utilitas Gambar

Fungsi Deskripsi
imageProxyToBitmap() Konversi ImageProxy (YUV_420_888 / JPEG) → Bitmap via CameraX built-in
resizeBitmap() Resize ke ukuran target (224×224)
rotateBitmap() Rotasi sesuai orientasi sensor kamera

4.3 Monitoring & Logging Layer

ResourceMonitor.kt

Metrik Cara Pengukuran
RAM (PSS) Debug.getMemoryInfo() → totalPss / 1024 → MB
CPU % Delta Process.getElapsedCpuTime() / Delta SystemClock.elapsedRealtime() × 100

CSVLogger.kt

  • Lokasi: Documents/leaf_detection_logs.csv (external files dir)
  • Kolom: Timestamp, Label, Confidence, Latency_ms, CPU_Usage, RAM_Usage_MB
  • Setiap frame real-time yang berhasil diklasifikasi → 1 baris CSV

5. Alur Data (Data Flow)

5.1 Mode Real-time (Kamera)

sequenceDiagram
    participant Camera as CameraX
    participant Exec as cameraExecutor
    participant BU as BitmapUtils
    participant ICH as ImageClassifierHelper
    participant UI as Main Thread (UI)
    participant Log as CSVLogger

    Camera->>Exec: ImageProxy (setiap frame)
    Exec->>BU: imageProxyToBitmap() + rotateBitmap()
    BU-->>Exec: Bitmap
    Exec->>UI: runOnUiThread
    UI->>ICH: classifyRealtime(bitmap)
    Note over ICH: Resize → Normalize → Quantize → Infer → Softmax
    ICH-->>UI: ClassificationResult
    UI->>UI: Update HUD overlay
    UI->>Log: log(result, cpu, ram)
Loading

5.2 Mode Snapshot (Capture / Gallery)

sequenceDiagram
    participant User as User
    participant CA as CameraActivity
    participant ICH as ImageClassifierHelper
    participant RA as ResultActivity

    User->>CA: Tap Capture / Pick Gallery
    CA->>CA: toBitmap() + rotateBitmap()
    CA->>ICH: classifyStatic(bitmap)
    ICH-->>CA: ClassificationResult
    CA->>CA: saveBitmapToCache()
    CA->>RA: Intent(label, confidence, latency)
    RA->>RA: displayResult()
Loading

6. Kelas Penyakit yang Dideteksi (13 Kelas)

# Label Tanaman
0 Potato___Early_blight 🥔 Kentang
1 Potato___Late_blight 🥔 Kentang
2 Potato___healthy 🥔 Kentang
3 Tomato__Target_Spot 🍅 Tomat
4 Tomato__YellowLeaf_Curl_Virus 🍅 Tomat
5 Tomato__Tomato_mosaic_virus 🍅 Tomat
6 Tomato_Bacterial_spot 🍅 Tomat
7 Tomato_Early_blight 🍅 Tomat
8 Tomato_Late_blight 🍅 Tomat
9 Tomato_Leaf_Mold 🍅 Tomat
10 Tomato_Septoria_leaf_spot 🍅 Tomat
11 Tomato_Spider_mites 🍅 Tomat
12 Tomato_healthy 🍅 Tomat

7. Technology Stack & Dependencies

graph LR
    subgraph "Core"
        K["Kotlin (JVM 17)"]
        AG["Android Gradle 36"]
    end

    subgraph "UI Framework"
        VB["ViewBinding (XML Layouts)"]
        MAT["Material Components"]
        COMPOSE["Jetpack Compose (theme only)"]
    end

    subgraph "Camera"
        CX["CameraX 1.4.0<br/>(camera-core, camera2,<br/>lifecycle, view)"]
    end

    subgraph "ML Runtime"
        TFL["TensorFlow Lite 2.17.0"]
        TFLS["TFLite Support 0.4.4"]
        LITERT["LiteRT API 1.0.1<br/>(substituted)"]
    end

    subgraph "Async"
        CR["Kotlin Coroutines 1.7.3"]
        LC["Lifecycle ViewModel/LiveData 2.7.0"]
    end

    K --> VB
    K --> CX
    K --> TFL
    K --> CR
    TFL -.->|"dependency substitution"| LITERT
Loading

Note

Terdapat dependency substitution di build.gradle.kts: tensorflow-lite-api di-redirect ke com.google.ai.edge.litert:litert-api:1.0.1 untuk kompatibilitas.

8. Status Pengembangan (Progress)

Tahap Status Keterangan
✅ Setup Gradle & Project Done Dependencies lengkap
✅ Model & Label Assets Done shufflenetv2_int8.tflite + labels.txt
✅ AndroidManifest Done Permissions, 2 Activities terdaftar
✅ Utility & Helpers Done BitmapUtils, ImageClassifierHelper, CSVLogger, ResourceMonitor sudah ada
✅ CameraActivity Done Kode ada, perlu testing di device
✅ ResultActivity Done Kode ada, perlu testing di device
✅ Resource Monitoring Done CSV logging sudah terimplementasi

9. Catatan Arsitektural

Important

Bug yang Sudah Diperbaiki (didokumentasikan langsung di kode):

  1. BUG KRITIS 1: bitmap selalu null di processImageAnalysis() karena finally { imageProxy.close() } dalam Kotlin membuang return value blok try. Solusi: pisahkan close dari try-catch.
  2. BUG KRITIS 2: lifecycleScope.launch{} dipanggil dari background thread → crash. Solusi: gunakan runOnUiThread {} sebelum launch.
  3. BUG MINOR: Capture/Gallery tidak menjalankan klasifikasi sebelum membuka ResultActivity. Solusi: tambahkan klasifikasi dan kirim hasil via Intent extras.

Tip

Peluang Improvement:

  • Migrasi dari Activity-based ke MVVM penuh dengan ViewModel + StateFlow
  • Compose theme sudah ada tapi belum digunakan — bisa migrasi UI ke full Compose
  • CSVLogger menulis file secara sinkron di setiap frame → pertimbangkan batching atau coroutine channel
  • classifyStatic() saat ini hanya memanggil classifyRealtime() — bisa dioptimasi khusus untuk gambar resolusi tinggi

About

No description, website, or topics provided.

Resources

Readme
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages