Intelligent Herbicide Applicator

We use computer vision to power agricultural tools that help farmers in developing countries apply herbicides more effectively and efficiently.

Due to inaccurate application, over 95% of herbicides reach a destination other than their target species, because they are sprayed or spread across entire agricultural fields.

Source: George Tyler Miller - Sustaining the Earth: an Integrated Approach

Our Story

We're four Berkeley data science grad students that have developed a a push cart dispenser which identifies weeds and dispenses appropriate herbicides to the targeted species.

Our Vision

Increase food security and help the environment by decreasing the overuse of herbicides by democratizing computer vision and machine learning to help farmers in developing countries efficiently and effectively apply herbicide to weeds while leaving desired crops alone.

Technology

We use deep learning, low cost Rasberry Pi hardware, and peristaltic pumps in a state-of-the-art device that takes pictures of weeds and plants, subsequently identifies the weeds, and finally applies the optimal amount of herbicides.

Who are we

The Power of Deep Learning

We've leveraged Convolutional Neural Networks to train a deep learning image classification model on images from our own farm. We've deployed our model to an edge device that we built to run in the real world.

Convolutional Neural Networks

Convolutional networks are neural networks that use convolutions in place of general matrix multiplications used in fully connected neural nets

Layers In CNN
- Convolution layer
- Pooling layer
- Fully connected layer

CNN Use Cases
- Image retrieval, detection
- Self driving cars
- Face/speech recognition,
- Text processing and so many others.

Image taken from Research Gate.

Animations taken from the Stanford deep learning tutorial.)

How it works

Architecture

Photos are taken and classified locally within the sprayer software ("edge computing") with a model developed in the cloud. Photos and classifications are sent to our cloud database where we provide in-depth analysis in our Enterprise App.

Hardware Device

How it works

Testimonials

Product Demo

Sprayer App Demo

Request demo

Futures

We plan to improve and expand the product and service the following ways:

Add a "plant-in-frame" recognizer, so operator needn't push a button
Add multiple language support to sprayer GUI (e.g. Arabic, French, others)
Improve system for collecting field images for further training (e.g. early season vs. late season images)
Develop hand-held version, with lighter LiPo battery
Add motor drive, row following, for autonomous/semi-autonomous operation
More distant future: Herbicide alternative (e.g. mechanical kill, selective laser heating of weeds)