Lensless Crop Disease Camera for Field Imaging

ISEF Category: Embedded Systems

Subcategory: Optics  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

A camera can work without a lens. That sounds impossible until you see how a diffuser and smart math can rebuild an image from a messy light pattern. For crop fields, that matters because cheap, small cameras can go where bulky optics cannot. Your project tests whether this trick can spot leaf disease outside the lab.

What Is It?

This project combines a lensless camera with on-device image reconstruction. Instead of a glass lens focusing light onto a sensor, a diffuser scatters the light into a coded pattern. Think of it like a scrambled fingerprint. The sensor records the scramble, and software works backward to recover the image.

The reconstruction part matters just as much as the hardware. An unrolled iterative network takes a math method like FISTA, which is a fast way to solve image recovery problems, and turns it into a neural network with fixed steps. That makes the system faster and easier to run on a microcontroller. In field use, the goal is not just to make a pretty picture. You want enough image quality to tell healthy leaves from diseased ones under changing sunlight and real-world motion.

Why This Is a Good Topic

This is a strong science fair topic because you can test real engineering choices, not just build a demo. You can compare diffuser types, reconstruction settings, lighting conditions, and disease classification accuracy. That gives you measurable results and a clear story about low-cost sensing for agriculture. A good version of this project also connects optics, embedded systems, and machine learning in a way that feels current and useful.

Research Questions

• How does diffuser pattern type affect reconstruction quality for leaf images? ?
• What is the effect of ambient field lighting on reconstruction accuracy? ?
• Does an unrolled FISTA network outperform classical iterative reconstruction on edge sharpness and disease visibility? ?
• To what extent does camera-to-leaf distance change classification accuracy for diseased versus healthy leaves? ?
• Which reconstruction setting gives the best tradeoff between speed and image fidelity on an MCU? ?
• How does motion blur from handheld capture affect disease detection performance? ?

Basic Materials

• OV5640 camera module.
• Microcontroller with enough memory and compute for edge inference.
• Diffuser material, such as translucent tape or diffuser film.
• Stable mounting frame or 3D-printed holder.
• Printed leaf targets or labeled crop leaf images.
• Notebook for capture conditions and ground truth labels.
• Phone or computer for data transfer and analysis.
• Controlled light source for indoor testing.

Advanced Materials

• OV5640 camera module with custom interface board.
• Microcontroller or embedded AI board with performance profiling tools.
• Multiple diffuser samples with known scattering properties.
• Optical bench or rigid alignment frame.
• Calibration target, such as a checkerboard or resolution chart.
• High-resolution reference camera for ground truth comparison.
• Spectrometer or light meter for illumination measurements.
• Compute access for training and testing reconstruction models.
• Environmental sensor for temperature, humidity, and light logging.

Software & Tools

• Python: Organizes image data, runs analysis, and compares reconstruction methods.
• ImageJ: Measures sharpness, contrast, and lesion visibility in reconstructed images.
• OpenCV: Handles image preprocessing, alignment, and simple computer vision tests.
• PyTorch: Trains or evaluates the unrolled reconstruction network.
• MATLAB: Supports signal processing, calibration, and quick algorithm checks if your school has access.

Experiment Steps

1. Define the imaging question you want to answer, such as reconstruction quality, classification accuracy, or speed on-device.
2. Choose one diffuser design and one reference capture setup so you can compare against a fixed baseline.
3. Plan a calibration workflow that links raw sensor output to reconstructed images and measurable quality scores.
4. Decide how you will test field conditions, including lighting changes, distance shifts, and motion from handheld use.
5. Select the metrics that matter most, such as sharpness, contrast, inference time, and disease detection accuracy.
6. Build controls that separate optical quality problems from algorithm problems, then compare your reconstruction methods fairly.

Common Pitfalls

• Using uncontrolled sunlight during every capture, which makes the sensor response change from one trial to the next.
• Skipping a clean calibration step, which leaves the reconstruction network unable to map raw patterns back to usable images.
• Comparing diseased and healthy leaves from different species, which confounds disease signal with plant identity.
• Judging image quality by eye only, which hides whether the system actually improves classification or contrast.
• Overloading the microcontroller with a model that is too large, which makes the system too slow for field use.

What Makes This Competitive

A competitive version of this project goes beyond making a working prototype. You would compare several diffuser designs, test more than one reconstruction approach, and report speed, accuracy, and power use side by side. You would also separate imaging quality from disease classification quality, since those are not always the same. Strong control experiments and careful statistics can make the project feel like real engineering research.

Project Variations

• Test the same lensless setup on tomato, pepper, or soybean leaves to see whether leaf shape changes reconstruction and disease detection.
• Swap the MCU for a different embedded board and compare speed, memory use, and accuracy on the same image set.
• Compare daytime outdoor captures with shaded or indoor captures to measure how field lighting changes performance.

Learn More

• PubMed: Search for review articles on lensless imaging, computational imaging, and plant disease detection to find recent methods and benchmarks.
• NASA Open Access papers: Search for computational imaging and sensor reconstruction papers that explain coded optics and inverse problems.
• MIT OpenCourseWare: Look for courses on signal processing, image processing, and computer vision to build the math background.
• USDA National Agricultural Library: Search for crop disease imaging studies and plant pathology references relevant to your target species.
• IEEE Xplore abstracts: Search for lensless cameras, DiffuserCam, and unrolled optimization papers, then read abstracts and open-access versions when available.