# app_rice_resnet50.py import streamlit as st import numpy as np import pandas as pd from PIL import Image import tensorflow as tf from tensorflow.keras.applications.resnet50 import preprocess_input # ----------------- SETTINGS ----------------- # Falls das Modell im Root liegt, nimm: "./best_rice_resnet50.keras" MODEL_PATH = "src/best_rice_resnet50.keras" IMG_SIZE = (224, 224) CLASS_NAMES = [ "Arborio", "Basmati", "Ipsala", "Jasmine", "Karacadag", ] # ----------------- MODEL LOADING ----------------- @st.cache_resource def load_resnet_model(): try: model = tf.keras.models.load_model(MODEL_PATH, compile=False) return model except Exception as e: st.error(f"Error loading model: {e}") return None def preprocess_image(image: Image.Image) -> np.ndarray: """Resize image and apply ResNet50 preprocessing.""" img = image.convert("RGB") img = img.resize(IMG_SIZE) img_array = np.array(img).astype("float32") img_array = preprocess_input(img_array) img_array = np.expand_dims(img_array, axis=0) return img_array def predict_image(model, image_array: np.ndarray): """Safe prediction with Top-3; robust gegen Längen-Mismatch.""" preds = model.predict(image_array, verbose=0)[0] probs = tf.nn.softmax(preds).numpy() n_model = len(probs) n_labels = len(CLASS_NAMES) n = min(n_model, n_labels) # Hauptklasse (nur im Bereich [0, n)) main_idx = int(np.argmax(probs[:n])) pred_class = CLASS_NAMES[main_idx] pred_conf = float(probs[main_idx]) # Top-k (hier 3) nur im gültigen Bereich k = min(3, n) topk_idx = np.argsort(probs[:n])[-k:][::-1] topk = [(CLASS_NAMES[i], float(probs[i])) for i in topk_idx] # Nur die ersten n Wahrscheinlichkeiten zurückgeben return pred_class, pred_conf, probs[:n], topk, n_model, n_labels # ----------------- PAGE CONFIG ----------------- st.set_page_config( page_title="Rice Classifier – ResNet50 (Transfer Learning)", page_icon="🌾", layout="wide", ) # ----------------- SIDEBAR ----------------- st.sidebar.title("🌾 Rice ResNet50 Classifier") st.sidebar.write( "This app uses a ResNet50 transfer learning model to classify rice grains." ) st.sidebar.subheader("Rice classes") for idx, name in enumerate(CLASS_NAMES): st.sidebar.write(f"- **{idx}** → {name}") st.sidebar.markdown("---") st.sidebar.caption("Author: Enes Yamac") # ----------------- MAIN PAGE ----------------- st.title("🌾 Rice Classifier – ResNet50 (Transfer Learning)") st.write("Upload a rice grain image. The model will predict the rice type and show top-3 results.") uploaded_file = st.file_uploader( "Upload a rice image (JPG or PNG)", type=["jpg", "jpeg", "png"], ) if uploaded_file is not None: image = Image.open(uploaded_file) col1, col2 = st.columns([1, 1]) with col1: st.subheader("Uploaded image") st.image(image, width=500) with st.spinner("Loading ResNet50 model and making prediction..."): model = load_resnet_model() if model is not None: img_array = preprocess_image(image) pred_class, pred_conf, probs, top3, n_model, n_labels = predict_image( model, img_array ) # kurze Warnung, falls Modell/Labels nicht exakt matchen if n_model != n_labels: st.warning( f"Model outputs {n_model} classes, CLASS_NAMES has {n_labels}. " f"Only the first {min(n_model, n_labels)} are shown." ) with col2: st.subheader("Prediction") st.markdown(f"**Predicted class:** `{pred_class}`") st.markdown(f"**Confidence:** `{pred_conf:.2%}`") st.subheader("Top-3 predictions") for name, p in top3: st.write(f"- {name}: {p:.2%}") st.subheader("Class probabilities") df = pd.DataFrame( {"class": CLASS_NAMES[: len(probs)], "probability": probs} ).set_index("class") st.bar_chart(df) with st.expander("Show probabilities as text"): for name, p in zip(CLASS_NAMES[: len(probs)], probs): st.write(f"- {name}: {p:.2%}") else: st.info("Please upload a rice image to get a prediction.")