import cv2
import numpy as np
import joblib
from skimage.feature import hog

svm_model = joblib.load("svm_model.pkl")
scaler = joblib.load("scaler.pkl")
pca = joblib.load("pca_model.pkl")
pca_scaler = joblib.load("pca_scaler.pkl")
label_mapping = np.load("label_mapping.npy", allow_pickle=True).item()

reverse_mapping = {v: k for k, v in label_mapping.items()}
pesticide_db = {
    "Tomato___Early_blight": "Chlorothalonil",
    "Tomato___Late_blight": "Mancozeb",
    "Tomato___healthy": "No pesticide required",
    "Maize___Common_rust": "Propiconazole",
    "Maize___healthy": "No pesticide required",
}
DEFAULT_PESTICIDE = "Consult Agricultural Expert"
def recommend_quantity(severity):
    if severity <= 10:
        return "No spray required (Monitoring stage)"
    elif severity <= 25:
        return "Low Dose (0.5 L/hectare)"
    elif severity <= 50:
        return "Medium Dose (1.0 L/hectare)"
    else:
        return "High Dose (1.5 L/hectare) - Immediate Action"

### Image Preprocessing
def preprocess_image(image_path, size=(128, 128)):
    img = cv2.imread(image_path)
    if img is None:
        raise ValueError("Image not found!")
    img = cv2.resize(img, size)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)    
    return gray, hsv

### Feature Extraction
def extract_features(gray_img, hsv_img):
    hog_features = hog(
        gray_img.astype('float32'),
        orientations=9,
        pixels_per_cell=(16, 16),
        cells_per_block=(2, 2),
        block_norm='L2-Hys',
        visualize=False,
        feature_vector=True
    ).astype('float32')  
    h_mean = np.mean(hsv_img[:, :, 0])
    s_mean = np.mean(hsv_img[:, :, 1])
    v_mean = np.mean(hsv_img[:, :, 2])
    color_features = np.array([h_mean, s_mean, v_mean], dtype='float32')
    combined = np.concatenate((hog_features, color_features))
    return combined.reshape(1, -1)

def calculate_severity(gray_img):
    _, leaf_mask = cv2.threshold(gray_img, 30, 255, cv2.THRESH_BINARY)
    _, disease_mask = cv2.threshold(
        gray_img, 0, 255,
        cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU
    )
    disease_mask = cv2.bitwise_and(disease_mask, leaf_mask)

    leaf_pixels = np.sum(leaf_mask == 255)
    infected_pixels = np.sum(disease_mask == 255)
    if leaf_pixels == 0:
        return 0
    severity = (infected_pixels / leaf_pixels) * 100
    return round(severity, 2)

def predict(image_path):
    gray, hsv = preprocess_image(image_path)
    features = extract_features(gray, hsv)
    features = pca_scaler.transform(features)
    features = pca.transform(features)
    features = scaler.transform(features)
    prediction = svm_model.predict(features)[0]
    disease_name = reverse_mapping[prediction]
    if "healthy" in disease_name.lower():
        severity = 0.0
    else:
        severity = calculate_severity(gray)
    if severity!=0.0:
        pesticide = pesticide_db.get(disease_name, DEFAULT_PESTICIDE)
    else:
        pesticide= "No consultant"
    quantity = recommend_quantity(severity)
    return disease_name,severity,pesticide,quantity

if __name__ == "__main__":
    image_path = "PlantVillage-Dataset/processed_dataset/Grape___healthy/3c593da4-e1df-460c-98a9-2bc71df670d5___Mt.N.V_HL 8970.JPG"  # Change to your test image 
    predict(image_path)