app.py

import tensorflow as tf
from PIL import Image
import numpy as np
import streamlit as st

@st.cache_resource
def load_model():
    model = tf.keras.models.load_model('/Users/shimodi/Documents/SolarDeepLearning/trained_effnet_hpo.keras')
    return model

def preprocess_image(image):
    img = image.resize((224, 224))
    img_array = np.array(img)
    img_array = np.expand_dims(img_array, axis=0)
    return img_array

def predict_defect(model, img):
    preprocessed = preprocess_image(img)
    prediction = model.predict(preprocessed)
    return prediction

st.set_page_config(
    page_title="Solar Panel Classifier",
    page_icon="🌞",
    layout="centered"
    )

st.title("Solar Panel Defect Detection App🌤️")
st.write("Upload an image of a solar panel to detect defects.")

uploaded_file = st.file_uploader("Choose an image...", type=["jpg", "jpeg", "png"])

if uploaded_file is not None:
    image = Image.open(uploaded_file).convert('RGB')
    st.image(image, caption="Uploaded Image", width="stretch")
    st.write("Analyzing...")

    model = load_model()
    prediction = predict_defect(model, image)
    classes = ['Bird-drop','Clean','Dusty','Electrical-damage','Physical-Damage','Snow-Covered']
    class_idx = np.argmax(prediction)
    defect = classes[class_idx]
    confidence = 100*np.max(prediction)
    st.subheader("Detection Result")
    if defect == "Clean":
        st.success("The panel appears to be in good condition!!")
    else:
        st.warning("A defect or contamination has been detected!!")
        st.write(f"Detected Defect: **{defect}** with Confidence: **{confidence:.2f}**%")

    
    # -------------------------------
    # Threshold-based probability highlighting
    # -------------------------------
    
    st.divider()
    st.subheader("Detailed Class Probabilities")
    
    threshold = st.slider(
        "Highlight classes above confidence (%)",
        min_value=10,
        max_value=90,
        value=50,
        step=5
    )
    
    probs = prediction[0] * 100  # convert to %

    predicted_class = classes[class_idx]
    
    for cls, prob in zip(classes, probs):
        label = f"{cls}: {prob:.2f}%"
        if cls == predicted_class:
            st.info(f"⭐ {label}")
        elif prob >= threshold:
            st.success(label)
        elif prob >= threshold / 2:
            st.warning(label)
        else:
            st.error(label)