Flora incognita

While smartphone cameras and social media have revolutionized many aspects of modern life, their potential impact on plant science remains largely untapped. This is where Flora Incognita steps in - a technological bridge connecting amateur nature enthusiasts with cutting-edge botanical research. Drawing from the combined expertise of Germany's Technical University of Ilmenau and Max Planck Institute for Biogeochemistry, this app has evolved into an AI-powered plant identification platform achieving 98.8% accuracy across 16,000 vascular plant species worldwide. The story of Flora Incognita is one of scientific collaboration, technological innovation, and citizen science coming together to transform how we document and understand the plant world.

App Overview

The Flora Incognita app has evolved significantly since its launch in 2018, expanding from an identification tool to a comprehensive plant documentation platform. Operating as a joint venture between the Technical University of Ilmenau and the Max Planck Institute for Biogeochemistry, the app has achieved remarkable success in plant species recognition, with a reported 98.8% identification accuracy across over 16,000 vascular plant species worldwide.

The app's development has overcome numerous technical challenges through innovative approaches in artificial neural network architecture. These networks, designed to mimic vertebrate brain structures, process image data through multiple hidden layers to recognize complex plant features. The identification process dynamically adapts to varying image conditions and user photographing habits, enhancing accuracy through situation-specific information requests.

Supported by extensive citizen science contributions, the app's database continues to grow through partnerships with educational institutions and research organizations. It now also supports offline mode functionality, allowing users to document plant observations without active internet connection. This feature, available in 20 languages across multiple device platforms, combines seamless user experience with sophisticated technological capabilities.

Plant Identification Process

The identification process begins when users take one or more images of specific plant organs, such as flowers or leaves, based on the unknown plant's growth form and current seasonal conditions. These images are captured using the smartphone or tablet camera and are processed through a cascade of deep neural networks hosted on the Flora Incognita computer cluster.

The neural networks operate in multiple hidden layers that work together to analyze and identify plant features autonomously. The process starts with raw data captured by input layer neurons, which are then passed through hidden layers before reaching the output layer. This hierarchical structure enables the network to recognize increasingly complex features, from simple geometric elements to complete flower and leaf structures.

To maintain high identification accuracy, the app employs several innovative techniques. These include systematic definition of complementary image perspectives for different plant groups, evaluation of habitat suitability information, and a situation-adaptive process that requests specific additional information based on the current identification certainty. This adaptive approach helps overcome challenges related to morphological variation within and between species, as well as variations in image quality and shooting conditions.

The app's training data comes from a large-scale citizen science project utilizing the companion Flora Capture app. This collaborative effort has been instrumental in building a comprehensive database capable of identifying over 5,000 German plant species with remarkable precision. The system has achieved a reported 98.8% identification accuracy across its extensive vascular plant species library, making it an invaluable tool for both amateur enthusiasts and botanical experts alike.

Development and Technology

This innovative plant identification technology evolved significantly from 2014 to 2019, combining smartphone cameras with advanced artificial neural networks inspired by vertebrate brain structures. The identification process operates across multiple hidden layers, autonomously analyzing and recognizing increasingly complex plant features from simple geometric elements to complete flower and leaf structures.

The development team addressed numerous technical challenges specific to plant identification, particularly the extensive morphological variation among German plant species. Drawing from a comprehensive citizen science initiative using the companion Flora Capture app, researchers developed several unique approaches to enhance identification accuracy. These included systematic perspective definition for different plant groups, habitat suitability evaluation, and a situation-adaptive process that dynamically requests additional information based on current identification certainty.

This adaptive approach proved particularly effective in recognizing over 5,000 German plant species, where traditional methods face significant challenges due to both intraspecific and interspecific variations. The system's deep learning capabilities enable it to process image data under diverse conditions, from different times of day and seasons to various species communities. Technical challenges related to user-specific photography habits and device-specific characteristics were also successfully addressed, contributing to the app's 98.8% identification accuracy across its extensive vascular plant species library.

User Features and Data Collection

The app's user interface has been designed with both ease-of-use and scientific rigor in mind. Each observation session begins with the selection of a specific plant organ, such as flower or leaf, based on the unknown plant's growth form and current seasonal conditions. Multiple images can be captured to provide different perspectives, with the app's deep neural networks analyzing each contribution simultaneously.

A particularly innovative feature of the app is its system of badges, which gamifies the plant documentation process while providing clear goals for users. Users can collect badges by documenting specific plant groups, with requirements typically involving the identification of 15 characteristic species within a particular community type. This approach combines popular gaming elements with rigorous botanical standards, encouraging both amateur enthusiasts and experienced botanists to participate in biodiversity monitoring.

All observations are timestamped and geolocated using precise GPS data, allowing researchers to track changes in plant distributions over time. The app's offline functionality enables users to capture and save observations when internet access is unavailable, with the option to identify these samples later when connectivity is restored.

The company behind Flora Incognita, the Max Planck Institute for Biogeochemistry, continues to expand both the app's capabilities and its scientific applications. Recent developments include expanded identification capabilities for invasive species, improved winter identification through leaf and bud analysis, and enhanced support for school-based citizen science projects.

As the app approaches its 10th millionth observation, the research team remains dedicated to improving identification accuracy and expanding the app's scientific applications. Future developments will focus on integrating additional plant attributes, such as pollinator-friendliness and invasive status, to provide users with comprehensive botanical information beyond simple identification.

Offline Mode Functionality

The app's offline mode enables users to document plant observations without active internet connection, storing information locally for processing later. Each offline observation requires capturing three images: a detailed view of the flower or leaf, a complete plant image, and a geographical reference shot. These images are saved to the device's photo gallery and remain accessible even when internet access is unavailable.

Upon reconnection, users can identify their stored observations through the app's standard identification process, which processes local image files alongside device metadata including timestamp and location information. The offline feature supports multiple language versions and cross-device profiles, allowing seamless observation and identification across different smart devices.

To help users document a wide range of plant types, the app supports three main observation modes: quick identification for simple cases, detailed observation for complex species, and guest profiles for one-time observations. The quick mode allows users to document common species with minimal interaction, while the detailed mode provides options for adding multiple perspectives and detailed notes. Guest profiles enable temporary observations without requiring a full user account, though these observations are cleared when the app is uninstalled or logged out.