About the system

Two signals read better than one

Most crop disease tools look only at a photo. But a leaf does not fall ill in isolation. Warm, wet weather is often what lets an infection take hold, so LeafGuard reads both the visual evidence in the leaf and the climate it grew in, then weighs them together.

What runs under the hood

The image branch uses MobileNetV2, a compact deep learning model pre-trained on millions of images and adapted here to tell healthy foliage from diseased. It was chosen because it stays light and fast enough to run on ordinary web hosting.

The weather branch reads two normalised values, temperature and rainfall. The two branches then join through feature fusion, so the final decision reflects the leaf and its environment at once. The trained model runs in the ONNX format for quick, lightweight inference.

Before anything is scanned

Every upload is first checked to confirm it actually shows a plant leaf. A selfie, a document, or a random object is turned away with a prompt to upload a clear leaf photo, so the model only ever reads what it was built to read.

What it does, and does not, do

• It gives a binary reading: Healthy or Diseased.
• It does not name the specific disease.
• It is a decision-support tool, not a substitute for an agronomist.
• Predictions are guidance, and will not always be correct.

Using it responsibly

Use LeafGuard to prompt earlier, closer inspection, not to make final treatment decisions on its own. Reliable field data and expert judgement remain essential to good farming.