Local Plant Medic

Devlog 6: Added Android building files to build APK artifact onto Github.

Devlog 5: Fixed code logic to understand wheat inputs better and predict diseases more accurately.

The issue was the elimination of lower-probability cases, causing empty inputs for all crops except for rice basically.

Devlog 4: Add Readme to describe the process of the build and the purpose of the project

Devlog 4: Cleared Cache for rebuild and added new analysis features successfully

Finally, after a few more tweaking and editing in the code as well as clearing the cache while testing on the vercel deployment, the app is ready to go! It features diagnostic summaries with condition analysis and symptom checkers. Furthermore, you are allowed to filter the plant for more accurate results. Below is a segment of the code featuring report diagnostics:

                  <div key={idx} className="p-4 border border-stone-200 rounded-xl bg-stone-50 space-y-2">
                    <div className="flex justify-between items-center">
                      <span className="text-lg font-bold text-stone-900 capitalize">{cleanLabel}</span>
                      <span className="text-xs font-bold text-emerald-800 bg-emerald-100 px-3 py-1 rounded-full">
                        {typeof p.score === 'number' ? `${(p.score * 100).toFixed(1)}%` : 'N/A'}
                      </span>
                    </div>
                    <p className="text-sm text-stone-600 italic">{info.desc}</p>
                    <p className="text-sm font-semibold text-stone-800">
                      Recommended Treatment: <span className="font-normal text-stone-600">{info.treat}</span>
                    </p>
                  </div>
                );```

Devlog 3: Refined the new ONNX and setup for new analysis features

After rollbacking a few broken commits, I made sure the code was ready to be setup for new analysis features such as giving advice for treating certain types of conditions. In addition, a future setup will be to allow users to narrow down the scope of analysis by stating the exact type of plant first.

Devlog 2: Fixing build errors and correct ONNX

I fixed the build errors by cleaning up the next.config.ts file and removing the curl commands that were causing the server to crash. I also updated the code in page.tsx so the image data would be sent as a Uint8Array instead of just an object. Finally, I fixed the worker.js file to wrap that data in a RawImage object, which let the AI model finally understand the images I was sending it. With these fixes, I have finally made the first version of the application that detects properly! However, much more work needs to be done when it comes to expanding analysis and the range of plants in context of agriculture.

Devlog 1: Initial setup

Built the first demo UI and connected the ONNX setup along with camera system. Much work to be done still as plant analysis needs to be properly trained locally!! As you can see values and analysis are a bit off since the local model isn’t trained very well yet.