predict.py

import os
import json
import librosa
import numpy as np
import tensorflow as tf
from tensorflow.keras.models import load_model
import matplotlib.pyplot as plt
from PIL import Image
from googleapiclient.discovery import build
from google.oauth2 import service_account
from googleapiclient.http import MediaIoBaseUpload
from flask import jsonify
import logging
from datetime import datetime
import subprocess
import tempfile
import shutil
from database import update_hive_health_in_db

# Ensure matplotlib runs in non-GUI mode
import matplotlib
matplotlib.use("Agg")

# Configure logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)

# Google Drive API Setup
SCOPES = ['https://www.googleapis.com/auth/drive.file']
PARENT_FOLDER_ID = os.getenv("GOOGLE_DRIVE_FOLDER_ID")

service_account_info = json.loads(os.getenv("G_Drive_Credentials"))
credentials = service_account.Credentials.from_service_account_info(service_account_info)

# Initialize Google Drive API
drive_service = build("drive", "v3", credentials=credentials)

# Load the bee/no bee model
MODEL_PATH = "./mobilenet_best_model_merged_bee_nobee.keras"
bee_model = load_model(MODEL_PATH)

# Load the queen model (MobileNet)
QUEEN_MODEL_PATH = "./mobilenet_best_model_merged_queen_noqueen.keras"
queen_model = load_model(QUEEN_MODEL_PATH)

# Load the mite attack model (MobileNet)
MITE_MODEL_PATH = "./mobilenet_best_model_merged_mite_nomite.keras"
mite_model = load_model(MITE_MODEL_PATH)

def check_ffmpeg():
    """Check if FFmpeg is available."""
    return shutil.which("ffmpeg") is not None

def convert_to_wav(input_path, output_path):
    """Convert audio file to WAV format using FFmpeg."""
    try:
        if not os.path.exists(input_path):
            raise FileNotFoundError("Input audio file does not exist")
        
        if not check_ffmpeg():
            raise RuntimeError("FFmpeg is not installed or not found in PATH")
        
        cmd = [
            "ffmpeg",
            "-i", input_path,
            "-acodec", "pcm_s16le",
            "-ar", "44100",
            "-ac", "2",
            "-y",
            output_path
        ]
        result = subprocess.run(
            cmd,
            stdout=subprocess.PIPE,
            stderr=subprocess.PIPE,
            text=True
        )
        if result.returncode != 0:
            raise RuntimeError(f"FFmpeg conversion failed: {result.stderr}")
        
        logger.info(f"Converted {input_path} to {output_path}")
        return output_path
    
    except Exception as e:
        logger.error(f"Error converting audio to WAV: {str(e)}")
        raise

def get_or_create_folder(folder_name, parent_id):
    """Retrieve or create a folder in Google Drive and return its ID."""
    query = f"name='{folder_name}' and mimeType='application/vnd.google-apps.folder' and '{parent_id}' in parents and trashed=false"
    response = drive_service.files().list(q=query, spaces='drive', fields='files(id, name)').execute()
    files = response.get('files', [])
    
    if files:
        return files[0].get('id')
    
    folder_metadata = {
        'name': folder_name,
        'mimeType': 'application/vnd.google-apps.folder',
        'parents': [parent_id]
    }
    folder = drive_service.files().create(body=folder_metadata, fields='id').execute()
    return folder.get('id')

def upload_to_drive(audio_file, result, user_predict=None):
    """Uploads an audio file to the appropriate Google Drive folder(s) based on prediction result and user_predict."""
    try:
        file_ids = []
        
        if result == "not bee":
            folder_name = "not bee"
            folder_id = get_or_create_folder(folder_name, PARENT_FOLDER_ID)
            file_metadata = {
                'name': audio_file.filename,
                'parents': [folder_id]
            }
            media = MediaIoBaseUpload(audio_file, mimetype=audio_file.content_type, resumable=True)
            file = drive_service.files().create(
                body=file_metadata,
                media_body=media,
                fields='id'
            ).execute()
            file_ids.append(file.get('id'))
            logger.info(f"Uploaded {audio_file.filename} to Google Drive folder '{folder_name}' with file ID: {file.get('id')}")
        
        else:
            bee_folder_id = get_or_create_folder("bee", PARENT_FOLDER_ID)
            audio_file.seek(0)
            file_metadata = {
                'name': audio_file.filename,
                'parents': [bee_folder_id]
            }
            media = MediaIoBaseUpload(audio_file, mimetype=audio_file.content_type, resumable=True)
            file = drive_service.files().create(
                body=file_metadata,
                media_body=media,
                fields='id'
            ).execute()
            file_ids.append(file.get('id'))
            logger.info(f"Uploaded {audio_file.filename} to Google Drive folder 'bee' with file ID: {file.get('id')}")
            
            if user_predict:
                user_predict = user_predict.strip().lower()
                valid_folders = {'healthy', 'no queen', 'mite attack', 'chalkbrood'}
                if user_predict in valid_folders:
                    folder_id = get_or_create_folder(user_predict, PARENT_FOLDER_ID)
                    audio_file.seek(0)
                    file_metadata = {
                        'name': audio_file.filename,
                        'parents': [folder_id]
                    }
                    media = MediaIoBaseUpload(audio_file, mimetype=audio_file.content_type, resumable=True)
                    file = drive_service.files().create(
                        body=file_metadata,
                        media_body=media,
                        fields='id'
                    ).execute()
                    file_ids.append(file.get('id'))
                    logger.info(f"Uploaded {audio_file.filename} to Google Drive folder '{user_predict}' with file ID: {file.get('id')}")
                else:
                    logger.warning(f"Ignoring invalid user_predict value: {user_predict}")
        
        return file_ids[0]
    
    except Exception as e:
        logger.error(f"Error uploading to Google Drive: {str(e)}")
        raise

def create_mel_spectrogram(audio_segment, sr):
    """Creates a mel spectrogram from an audio segment."""
    try:
        spectrogram = librosa.feature.melspectrogram(y=audio_segment, sr=sr, n_mels=128)
        spectrogram_db = librosa.power_to_db(spectrogram, ref=np.max)

        plt.figure(figsize=(2, 2), dpi=100)
        plt.axis('off')
        plt.imshow(spectrogram_db, aspect='auto', cmap='magma', origin='lower')
        plt.tight_layout(pad=0)
        
        temp_image_path = "/tmp/temp_spectrogram.png"
        plt.savefig(temp_image_path, bbox_inches='tight', pad_inches=0)
        plt.close()

        img = Image.open(temp_image_path).convert('RGB')
        img = img.resize((224, 224))
        img_array = np.array(img) / 255.0

        os.remove(temp_image_path)
        return img_array

    except Exception as e:
        logger.error(f"Error creating spectrogram: {e}")
        return None

def predict_queen_audio(file_path, model):
    """Processes audio for queen detection using 10-second segments."""
    try:
        y, sr = librosa.load(file_path, sr=None)
        duration = librosa.get_duration(y=y, sr=sr)
        if duration <= 10:
            return {"error": "Audio file must be longer than 10 seconds"}

        healthy_count = 0
        total_segments = 0
        segment_start = 0

        while segment_start < duration:
            segment_end = min(segment_start + 10, duration)
            if segment_end - segment_start < 10 and segment_start > 0:
                segment_start = max(0, duration - 10)
                segment_end = duration

            audio_segment = y[int(segment_start * sr):int(segment_end * sr)]
            spectrogram = create_mel_spectrogram(audio_segment, sr)

            if spectrogram is not None:
                spectrogram = np.expand_dims(spectrogram, axis=0)
                prediction = model.predict(spectrogram)
                probability = prediction[0][0]  # Assuming binary classification (0: no queen, 1: healthy)
                if probability >= 0.8:  # Aligned with predict_audio for consistency
                    healthy_count += 1
                total_segments += 1

            segment_start += 10

        if total_segments > 0:
            healthy_percentage = (healthy_count / total_segments) * 100
            result = "healthy" if healthy_percentage >= 70 else "no queen"
            return result
        else:
            return {"error": "No valid segments processed"}

    except Exception as e:
        logger.error(f"Error in queen prediction: {e}")
        return {"error": str(e)}

def predict_mite_audio(file_path, model):
    """Processes audio for mite attack detection using 10-second segments."""
    try:
        y, sr = librosa.load(file_path, sr=None)
        duration = librosa.get_duration(y=y, sr=sr)
        if duration <= 10:
            return {"error": "Audio file must be longer than 10 seconds"}

        healthy_count = 0
        total_segments = 0
        segment_start = 0

        while segment_start < duration:
            segment_end = min(segment_start + 10, duration)
            if segment_end - segment_start < 10 and segment_start > 0:
                segment_start = max(0, duration - 10)
                segment_end = duration

            audio_segment = y[int(segment_start * sr):int(segment_end * sr)]
            spectrogram = create_mel_spectrogram(audio_segment, sr)

            if spectrogram is not None:
                spectrogram = np.expand_dims(spectrogram, axis=0)
                prediction = model.predict(spectrogram)
                probability = prediction[0][0]  # Assuming binary classification (0: mite attack, 1: healthy)
                if probability >= 0.8:  # Consistent threshold
                    healthy_count += 1
                total_segments += 1

            segment_start += 10

        if total_segments > 0:
            healthy_percentage = (healthy_count / total_segments) * 100
            result = "healthy" if healthy_percentage >= 70 else "mite attack"
            return result
        else:
            return {"error": "No valid segments processed"}

    except Exception as e:
        logger.error(f"Error in mite attack prediction: {e}")
        return {"error": str(e)}

def predict_audio(audio_path, request_id):
    """Predicts whether an audio file contains bee sounds."""
    try:
        y, sr = librosa.load(audio_path, sr=None)
        if y is None or sr is None:
            return {"error": "Failed to load audio"}

        duration = librosa.get_duration(y=y, sr=sr)
        if duration <= 10:
            return {"error": "Audio file must be longer than 10 seconds"}

        bee_count = 0
        total_segments = 0
        segment_start = 0

        while segment_start < duration:
            segment_end = min(segment_start + 10, duration)
            if segment_end - segment_start < 10 and segment_start > 0:
                segment_start = max(0, duration - 10)
                segment_end = duration

            audio_segment = y[int(segment_start * sr):int(segment_end * sr)]
            spectrogram = create_mel_spectrogram(audio_segment, sr)

            if spectrogram is not None:
                spectrogram = np.expand_dims(spectrogram, axis=0)
                prediction = bee_model.predict(spectrogram)
                probability = prediction[0][0]

                if probability <= 0.2:
                    bee_count += 1

                total_segments += 1

            segment_start += 10

        if total_segments > 0:
            bee_percentage = (bee_count / total_segments) * 100
            result = "bee" if bee_percentage >= 70 else "not bee"
            logger.info(f"Request {request_id} - Prediction result: {result}")
            return {"result": result}
        else:
            return {"result": "try again"}

    except Exception as e:
        logger.error(f"Request {request_id} - Error during bee prediction: {e}")
        return {"error": str(e)}

def handle_predict(request, save_prediction):
    """Handles the prediction logic for the /predict route."""
    request_id = datetime.now().strftime("%Y%m%d_%H%M%S_%f")
    logger.info(f"Request {request_id} - Starting prediction")

    if 'audio' not in request.files or 'user_id' not in request.form:
        logger.error(f"Request {request_id} - Missing data")
        return jsonify({"error": "Missing data"}), 400

    user_id = request.form['user_id']
    audio_file = request.files['audio']
    hive_id = request.form.get('hive_id')
    user_predict = request.form.get('user_predict')
    
    if audio_file.filename == '':
        logger.error(f"Request {request_id} - No file selected")
        return jsonify({"error": "No file selected"}), 400

    if not check_ffmpeg():
        logger.error(f"Request {request_id} - FFmpeg not found")
        return jsonify({"error": "Server error: FFmpeg is not available"}), 500

    temp_dir = tempfile.gettempdir()
    original_filename = audio_file.filename
    original_ext = os.path.splitext(original_filename)[1].lower()
    temp_original_path = os.path.join(temp_dir, f"temp_audio_{request_id}{original_ext}")
    audio_file.save(temp_original_path)

    temp_wav_path = os.path.join(temp_dir, f"temp_audio_{request_id}.wav")
    if original_ext not in ['.wav', '.mp3']:
        try:
            convert_to_wav(temp_original_path, temp_wav_path)
        except Exception as e:
            os.remove(temp_original_path)
            logger.error(f"Request {request_id} - Audio conversion failed: {str(e)}")
            return jsonify({"error": f"Failed to process audio file: {str(e)}"}), 400
    else:
        temp_wav_path = temp_original_path
        logger.info(f"Request {request_id} - Using original file (no conversion needed): {temp_wav_path}")

    prediction_result = predict_audio(temp_wav_path, request_id)
    
    if "error" in prediction_result:
        os.remove(temp_original_path)
        if temp_wav_path != temp_original_path and os.path.exists(temp_wav_path):
            os.remove(temp_wav_path)
        logger.error(f"Request {request_id} - Prediction failed: {prediction_result['error']}")
        return jsonify({"result": "try again"}), 400

    result = prediction_result["result"].lower()

    if result == "try again":
        os.remove(temp_original_path)
        if temp_wav_path != temp_original_path and os.path.exists(temp_wav_path):
            os.remove(temp_wav_path)
        logger.info(f"Request {request_id} - Result: try again")
        return jsonify({"result": "try again"})

    if result == "bee":
        queen_result = predict_queen_audio(temp_wav_path, queen_model)
        mite_result = predict_mite_audio(temp_wav_path, mite_model)
        
        if "error" in queen_result or "error" in mite_result:
            os.remove(temp_original_path)
            if temp_wav_path != temp_original_path and os.path.exists(temp_wav_path):
                os.remove(temp_wav_path)
            logger.error(f"Request {request_id} - Queen prediction failed: {queen_result.get('error', 'Unknown')}, Mite prediction failed: {mite_result.get('error', 'Unknown')}")
            return jsonify({"result": "try again"}), 400

        logger.info(f"Request {request_id} - Queen prediction result: {queen_result}, Mite prediction result: {mite_result}")

        # Combine results according to specified logic
        if queen_result == "healthy" and mite_result == "healthy":
            result = "healthy"
        elif queen_result == "no queen" and mite_result == "healthy":
            result = "no queen"
        elif queen_result == "healthy" and mite_result == "mite attack":
            result = "mite attack"
        elif queen_result == "no queen" and mite_result == "mite attack":
            result = "no queen,mite attack"
        else:
            result = "try again"  # Fallback for unexpected cases
            logger.warning(f"Request {request_id} - Unexpected combination: queen={queen_result}, mite={mite_result}")

    if user_predict and hive_id:
        try:
            user_predict = user_predict.strip().lower()
            update_hive_health_in_db(hive_id, user_predict)
            logger.info(f"Request {request_id} - Updated hive {hive_id} health_status to {user_predict}")
        except Exception as e:
            logger.error(f"Request {request_id} - Failed to update hive health status: {str(e)}")
            return jsonify({"error": f"Failed to update hive health status: {str(e)}"}), 400

    file_id = None
    try:
        audio_file.seek(0)
        file_id = upload_to_drive(audio_file, result, user_predict)
    except Exception as e:
        logger.error(f"Request {request_id} - Failed to upload to Google Drive: {str(e)}")
        # Continue with saving prediction and returning result with file_id=None

    save_prediction(user_id, audio_file.filename, result, file_id, hive_id, user_predict)

    os.remove(temp_original_path)
    if temp_wav_path != temp_original_path and os.path.exists(temp_wav_path):
        os.remove(temp_wav_path)

    logger.info(f"Request {request_id} - Final result: {result}, file_id: {file_id}")
    return jsonify({"result": result, "file_id": file_id})