Here’s a complete English guide for building a wheat protein estimation device using Raspberry Pi + Arduino, based on visible/near-infrared (Vis/NIR) spectroscopy.

---

🌾 1) Concept Overview

Protein content in wheat kernels correlates with light absorption in the near-infrared region (NIR).
We measure reflected or transmitted light at several wavelengths and use a regression model (PLS) to estimate protein percentage.

🧠 System flow:
LEDs → Sample → Spectral Sensor → Arduino (data capture) → Raspberry Pi (data processing + display)

⚠️ Note: This is a calibration-based monitoring device, not a lab-accurate analyzer. Still, it’s great for real-time field or combine harvester use.

---

🔧 2) Hardware Components

Option A – Basic Prototype

• Raspberry Pi 4 (or Zero 2 W)
• Arduino Nano / Pro Mini (optional, for fast sensor reads)
• AS7341 multispectral sensor (visible + near IR, ~20€)
• White + NIR LEDs (940 nm)
• Tube enclosure with matte black interior
• DC-DC 12→5 V converter and I²C wiring

Option B – Recommended Field Version

• AS7265x “Triad” sensor (18 bands, 410–940 nm, ~80–100 €)
• NIR LEDs (940/1050/1200 nm) with constant current driver
• Simple in-line grain chute with quartz window
• Optional NIR photodiode for reference channel

Option C – Semi-Pro

• Mini spectrometer (e.g., Hamamatsu C12880MA)
• Diffraction optics or fiber input
• Stable NIR lamp source
• Higher precision, higher cost

---

⚙️ 3) Optical & Mechanical Setup

• LED at ~45°, sensor near perpendicular → diffuse reflection geometry.
• Paint interior black (to reduce stray light).
• Keep sample layer thickness consistent.
• Add temperature sensor for compensation.
• Protect window from dust (compressed air or brush).

---

🔌 4) Wiring (Example for AS7265x via I²C)

• VCC → 3.3 V
• GND → GND
• SDA → GPIO2
• SCL → GPIO3
• LEDs → current driver, controlled via GPIO or Arduino pin
• Arduino → USB serial connection to Pi

---

💻 5) Software Flow

Arduino (Sensor Reader)

include

include “AS7265X.h”

AS7265X sensor;

void setup() {
Serial.begin(115200);
Wire.begin();
if(!sensor.begin()) { Serial.println(“No sensor found”); while(1); }
sensor.setMeasurementMode(AS7265X_MEASUREMENT_MODE_6CHAN_CONTINUOUS);
sensor.setIntegrationTime(50);
sensor.setGain(AS7265X_GAIN_64X);
}

void loop() {
sensor.takeMeasurements();
Serial.print("{\"bands\":[");
for(int i=0;i<18;i++){
Serial.print(sensor.getCalibratedValue(i),3);
if(i<17) Serial.print(",");
}
Serial.println("]}");
delay(100);
}

Raspberry Pi (Python Data Processing)

import serial, json, time, numpy as np, joblib
ser = serial.Serial('/dev/ttyACM0', 115200, timeout=1)
pls = joblib.load('pls_model.pkl')
scaler = joblib.load('scaler.pkl')

while True:
    line = ser.readline().decode('utf-8', 'ignore').strip()
    if not line: continue
    try:
        x = np.array(json.loads(line)['bands']).reshape(1, -1)
        x = scaler.transform(x)
        y = pls.predict(x)[0, 0]
        print(f"{time.strftime('%H:%M:%S')}  Protein: {y:.2f}%")
    except Exception:
        pass

6) Calibration – The Key Step

1. Collect grain samples during harvest under varied conditions (varieties, moisture).
2. For each sample:
   • Record 100–200 spectra (average them).
   • Get lab reference protein values (Kjeldahl/NIR).
3. Combine data:
   • X = spectra; y = lab protein %.
4. Preprocess:
   • Apply SNV or 1st derivative (Savitzky–Golay).
5. Train PLSRegression model (choose components via cross-validation).
6. Validate with test set, report R² and RMSEP.
7. Save model (pls_model.pkl) and scaler (scaler.pkl) for deployment.

🌡️ Include grain moisture as a feature or ensure calibration covers similar moisture range.

---

🚜 7) Mounting on Combine

• Best location: clean grain elevator or outlet, where grain flows evenly.
• Use anti-vibration mount.
• Dust protection: glass window + low airflow cleaning.
• Power via 12 V→5 V converter.
• Keep LEDs/sensor cool.

---

💰 8) Cost Estimate (Option B)

Component	Approx. Cost (€)
Raspberry Pi 4	70
Arduino Nano	8
AS7265x sensor	90
NIR LEDs + driver	30
Enclosure / 3D print	40
Power + wiring	30
Total	~250 €