import streamlit as st import numpy as np import cv2 from PIL import Image from io import BytesIO from ultralytics import YOLO import os import tempfile import base64 import requests from datetime import datetime from gtts import gTTS from googletrans import Translator import google.generativeai as genai # Import Gemini API # Configure Google Gemini API GEMINI_API_KEY = os.getenv("GOOGLE_API_KEY") genai.configure(api_key=GEMINI_API_KEY) # Load YOLO model for crop disease detection yolo_model = YOLO("models/best.pt") # Initialize conversation history if not set if "conversation_history" not in st.session_state: st.session_state.conversation_history = {} # Function to preprocess images def preprocess_image(image, target_size=(224, 224)): """Resize image for AI models.""" image = Image.fromarray(image) image = image.resize(target_size) return image # Generate response from Gemini AI with history def generate_gemini_response(disease_list, user_context="", conversation_history=None): """Generate a structured diagnosis using Gemini API, considering conversation history.""" try: model = genai.GenerativeModel("gemini-1.5-pro") # Start with detected diseases prompt = f""" You are an expert plant pathologist. The detected crop diseases are: {', '.join(disease_list)}. User's context or question: {user_context if user_context else "Provide a general analysis"} """ # Add past conversation history for better continuity if conversation_history: history_text = "\n\nPrevious conversation:\n" for entry in conversation_history: history_text += f"- User: {entry['question']}\n- AI: {entry['response']}\n" prompt += history_text # Ask Gemini for a structured diagnosis prompt += """ Provide a detailed diagnosis including: 1. Symptoms 2. Causes and risk factors 3. Impact on crops 4. Treatment options (short-term & long-term) 5. Prevention strategies """ response = model.generate_content(prompt) return response.text if response else "No response from Gemini." except Exception as e: return f"Error connecting to Gemini API: {str(e)}" # Perform inference using YOLO def inference(image): """Detect crop diseases in the given image.""" results = yolo_model(image, conf=0.4) infer = np.zeros(image.shape, dtype=np.uint8) detected_classes = [] class_names = {} for r in results: infer = r.plot() class_names = r.names detected_classes = r.boxes.cls.tolist() return infer, detected_classes, class_names # Convert text to speech def text_to_speech(text, language="en"): """Convert text to speech using gTTS.""" try: with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as temp_audio: tts = gTTS(text=text, lang=language, slow=False) tts.save(temp_audio.name) with open(temp_audio.name, "rb") as audio_file: audio_bytes = audio_file.read() os.unlink(temp_audio.name) return audio_bytes except Exception as e: st.error(f"Error generating speech: {str(e)}") return None # Initialize Streamlit UI st.title("🌱 AI-Powered Crop Disease Detection & Diagnosis 🔬") # Sidebar settings with st.sidebar: st.header("Settings") # Fake model selection (Still uses Gemini) selected_model = st.selectbox("Choose Model", ["Gemini", "GPT-4", "Claude", "Llama 3", "Mistral"], help="This app always uses Gemini.") confidence_threshold = st.slider("Detection Confidence Threshold", 0.0, 1.0, 0.4) # Text-to-Speech Settings tts_enabled = st.checkbox("Enable Text-to-Speech", value=True) language = st.selectbox("Speech Language", options=["en", "es", "fr", "de"], format_func=lambda x: { "en": "English", "es": "Spanish", "fr": "French", "de": "German" }[x]) if st.button("Clear Conversation History"): st.session_state.conversation_history = {} st.success("Conversation history cleared!") # User context input st.subheader("📝 Provide Initial Context or Ask a Question") user_context = st.text_area("Enter any details, symptoms, or questions about the plant's condition.", placeholder="Example: My tomato plant leaves are turning yellow. Is it a disease or a nutrient deficiency?") # Upload an image uploaded_file = st.file_uploader("📤 Upload a plant image", type=["jpg", "jpeg", "png"]) if uploaded_file: file_id = uploaded_file.name # Initialize conversation history for this image if not set if file_id not in st.session_state.conversation_history: st.session_state.conversation_history[file_id] = [] # Convert file to image file_bytes = np.asarray(bytearray(uploaded_file.read()), dtype=np.uint8) img = cv2.imdecode(file_bytes, 1) # Perform inference processed_image, detected_classes, class_names = inference(img) # Display processed image with detected diseases st.image(processed_image, caption="🔍 Detected Diseases", use_column_width=True) if detected_classes: detected_disease_names = [class_names[cls] for cls in detected_classes] st.write(f"✅ **Detected Diseases:** {', '.join(detected_disease_names)}") # AI-generated diagnosis from Gemini st.subheader("📋 AI Diagnosis") with st.spinner("Generating diagnosis... 🔄"): diagnosis = generate_gemini_response(detected_disease_names, user_context, st.session_state.conversation_history[file_id]) # Save response to history st.session_state.conversation_history[file_id].append({"question": user_context, "response": diagnosis}) # Display the diagnosis st.write(diagnosis) # Show past conversation history if st.session_state.conversation_history[file_id]: st.subheader("🗂️ Conversation History") for i, entry in enumerate(st.session_state.conversation_history[file_id]): with st.expander(f"Q{i+1}: {entry['question'][:50]}..."): st.write("**User:**", entry["question"]) st.write("**AI:**", entry["response"]) # Convert diagnosis to speech if enabled if tts_enabled: if st.button("🔊 Listen to Diagnosis"): with st.spinner("Generating audio... 🎵"): audio_bytes = text_to_speech(diagnosis, language) if audio_bytes: st.audio(audio_bytes, format="audio/mp3") else: st.write("❌ No crop disease detected.") # Instructions for users st.markdown(""" --- ### How to Use: 1. Upload an image of a plant leaf with suspected disease. 2. Provide context (optional) about symptoms or concerns. 3. The system detects the disease using AI. 4. Gemini generates a diagnosis with symptoms and treatments. 5. Ask follow-up questions, and the AI will remember previous responses. 6. Optionally, listen to the AI-generated diagnosis. """)