berkano.wladradchenko.ru

<div style="display: flex; position: relative; flex-direction: row;"> <a href="https://github.com/wladradchenko/berkano.wladradchenko.ru/blob/main/LICENSE"> <img src="https://img.shields.io/badge/license-MIT-yellow.svg" alt="License: MIT v1.0" style="margin: 5px;"/> </a> <a href="https://github.com/wladradchenko/berkano.wladradchenko.ru"> <img src="https://img.shields.io/badge/GitHub-24292e?style=flat&logo=github&logoColor=yellow" alt="GitHub" style="margin: 5px;"/> </a> </div> <div id="top"></div> <br /> <div align="center"> <a href="https://github.com/wladradchenko/berkano.wladradchenko.ru"> <img src="https://huggingface.co/wladradchenko/berkano.wladradchenko.ru/resolve/main/assets/leaves.svg" alt="Logo" width="180" height="180"> </a> <h3 align="center">Berkano: Mobile Models for Plant Analysis</h3> <p align="center"> <a href="https://github.com/wladradchenko/berkano.wladradchenko.ru/wiki">Documentation</a> <br/> <a href="https://github.com/wladradchenko/berkano.wladradchenko.ru/issues">Issue</a> · <a href="https://github.com/wladradchenko/berkano.wladradchenko.ru/discussions">Discussions</a> </p> </div> <!-- ABOUT THE PROJECT -->

About

Plant Analysis Models (Mobile-Optimized ONNX). This repository contains a collection of ONNX models for plant image analysis, optimized for mobile devices (Android / iOS) as well as for standard machine learning workflows. The models are designed for fast inference and can be executed locally on-device or in server-side environments.

The repository is structured to be easily integrated into mobile applications or used as a ready-to-use model bundle for research and prototyping.

Repository structure

The repository consists of two main directories: models and files.

The models directory contains ONNX models, each solving a specific plant analysis task. All models are optimized for mobile inference and are compatible with ONNX Runtime, including onnxruntime-react-native.

The files directory contains auxiliary data required to correctly interpret the outputs of certain models.

Models

disease_detections.onnx is a plant disease detection model based on image embeddings. Instead of performing direct multi-class classification, the model extracts an embedding from the input image and performs similarity search against a precomputed set of disease embeddings stored in embeddings.bin. This approach is significantly faster and more efficient on mobile devices compared to classical classification. The model is trained on the LeafNet dataset and is suitable for offline and real-time mobile usage.

plant_analysis.onnx is a regression model designed to estimate plant age and leaf count from a single image. The model takes a plant image as input and outputs numerical values representing the predicted age and number of leaves. It is trained using a MobileNet-based architecture and optimized for execution on mobile devices.

plant_classifications.onnx is a plant species classification model trained on the PlantNet dataset. It is intended for recognizing wild and exotic plant species. The model is not designed for household or agricultural crops and may produce unreliable results when used outside its training domain.

Auxiliary files

class_mapping.txt provides a mapping from the output class indices of the plant_classifications.onnx model to plant species identifiers.

species_id_to_name.txt contains human-readable plant names corresponding to each species identifier. These two files are used together to convert raw model outputs into meaningful plant names.

embeddings.bin stores vector representations of plant diseases used by the disease_detections.onnx model for similarity-based disease identification.

captions.json maps embedding identifiers to disease names, allowing the nearest embedding match to be translated into a textual disease description.

Usage

The models can be used in Python-based machine learning projects as well as in mobile applications using ONNX Runtime. All models are provided in ONNX format and do not require additional conversion for inference.

A reference implementation demonstrating both ML and mobile usage is available in a separate GitHub repository. A link to this repository is provided in the project description.

Purpose and limitations

These models are intended for research, educational, and applied plant analysis tasks. The outputs are not intended to serve as medical, agricultural, or professional diagnostic advice and should not be used as the sole basis for critical decision-making.

<!-- DONAT -->

Support the Project

You can support the author of the project in the development of his creative ideas on <a href="https://boosty.to/wunjo">Boosty</a> or <a href="https://wladradchenko.ru/donat">CloudTips</a>.

<div align="center"> <table> <tr> <th>Boosty</th> <th>CloudTips</th> </tr> <tr align="center"> <td><img src="https://raw.githubusercontent.com/wladradchenko/wunjo.wladradchenko.ru/main/example/wunjo-donate.png" alt="pizza" width="250"></td> <td><img src="https://github.com/wladradchenko/wunjo.wladradchenko.ru/assets/56233697/acc80acd-0e39-4476-88db-0a10f2098e25" alt="coffee" width="250"></td> </tr> </table> </div> <!-- CONTACT -->

Contact

Owner: Wladislav Radchenko

Email: [email protected]

Project on GitHub: https://github.com/wladradchenko/berkano.wladradchenko.ru

<p align="right">(<a href="#top">to top</a>)</p>