Updated Apr 04, 2018

Default inno

LeafAI

LeafAI, a machine learning based plant disease identification and management platform for urban gardeners and subsistence farmers, is able to accurately and quickly classify low quality images taken from a smartphone through a mobile and web app.

leafai.org
Default avatar

Maanasa Mendu

Send Message

Stage 3: Proof of Concept

The mobile and web based application have been tested on images representative of field conditions (with background clutter, varying angles, scales, and amounts of illumination) and have obtained an accuracy of greater than 90%.

Registered in United States.

Focus Areas:

Agriculture, Crowdsourcing and Open Source and Technology

Agriculture, Crowdsourcing and Open Source and TechnologySEE LESS

Implemented In:

United States

United StatesSEE LESS

1
Country Implemented In
2
Employee
$1,000
Funds Raised to Date

Problem

Nearly 700,000 people around the world suffer from malnutrition. Alongside this, nearly 75% of our total crop output comes from just 12 plant species. The homogeneity of the current agriculture system combined with the effects of climate change has led to a growing threat - plant disease. The key to reducing the damage of plant disease lies in early and accurate detection. Current plant disease diagnostic tools like lab methods (ELISA and PCR) are time consuming and expensive and the visual inspection of plants by plant pathologists is often a rare resource in developing countries. We need an inexpensive, widespread, and fast plant disease identification tool.

}

Solution

LeafAI utilizes state-of-the-art technology - convolutional neural networks - to diagnose plant disease with an overall accuracy of 95.6% and under 1 second on the Plant Village Dataset. Without any feature extraction, LeafAI is able to classify images of 38 classes of agronomic crops (such as soybeans and corn) with and without disease. LeafAI is able to classify imperfect images - low resolution, varying angles, and scales. This concept is scalable as the neural networks may be continuously trained to identify upcoming diseases.

Target Beneficiaries

We aim to serve both urban gardeners and subsistence farmers. In developing countries, for subsistence and small holder farmers, famring is more than an economic means; it's a way life. By 2020 there will be over 5 billion smartphone subscriptions worldwide making LeafAI potentially a universal diagnostic tool. Whether integrated as an mobile application coupled with a web server for individual uses or even in the form of a continuous drone based monitoring system, LeafAI can provide an inexpensive and accurate diagnostic tool. Even though LeafAI does not solve the plant disease problem, it makes us one step closer towards providing economic and food security to all.

Mission and Vision

We hope to bring the world closer towards economic and food security through this automated fast and accurate plant diseases diagnostic tool. LeafAI does not only identify the plant disease but also provides information about the treatment, causes, and symptoms of the said disease. One of the key features of the app is the "alerting system" for farmers based on crowd sourced geo-spatial data gathered from the users of the app. Nearby farmers will be notified of an outbreak within their area. It's important to note, however, that LeafAI is a supplementary tool. In the app, a directory of local resources - aid organizations and plant clinics - is available as a resources for users to obtain further access to experts.

Innovation Description

The mobile application of LeafAI is made possible through the PlantNet algorithm and web server. The PlantNet algorithm is a convolutional neural network (CNN) trained on the Plant Village and Bugwood dataset to identify 38 classes of agronomic crops with and without diseases. To develop this optimal algorithm, we tested 9 different variations with different network architectures, datasets, and methods of training. The neural network takes in a raw image and converts it through a series of hidden layers which apply filters and downsizing to reduce the complex image data into something which can be manipulated and assigned class probabilities.

Competitive Advantage

Existing plant disease diagnostic tools like ELISA and PCR are time consuming and expensive ($200-300 per unit) and the visual inspection of plant diseases by plant pathologists is often a rare resource in developing countries. Even existing digital tools such as the CABI crop compendium, Purdue Plant Doctor App, and the University of Hawaii's LeafDoctor rely on manual analysis. LeafAI instead provides a fast, automated, and accurate plant disease identification system based on symptoms such as abnormal coloration and wilting that appear on the leaves of plants to classify biotic plant diseases. Machine Learning and plant disease have been tried together; however, machine learning based approaches of today rely on a feature extraction where the photo taken by the user undergoes a series of pre-processing steps which are time consuming and make the network non generalizable. Overall, the LeafAI solution is inexpensive, only requiring a phone with a capability to take an image of any quality.

Planned Goals and Milestones

We're enhancing the accuracy of LeafAI and reducing the amount of overfitting through training with a more diverse dataset, utilizing image segmentation to remove background clutter, and even expanding the system's capability to recognize symptoms that appear on the shoots, fruits, and stems of plants. Alongside this, we are collaborating with researchers and plant pathologists to obtain a dataset representative of field conditions and developing diseases.
Funding Goal5,000
New FeatureInformation regarding each individual disease's symptoms, treatments, and causes. Plant pathologist near me type directory.

Milestones

Jan 2018
New Product or Service
Launched the mobile application for LeafAI
Dec 2017
New Product or Service
Launched the web application for LeafAI
Sep 2017
Created
Date Unknown
New Country Implemented In
United States