SBIR Phase I: Use of Machine Learning Techniques for Robust Crop and Weed Detection in Agricultural Fields

Blue River Technology Inc, Sunnyvale CA

Investigators

Abstract

This Small Business Innovation Research (SBIR) Phase I project seeks to understand the fundamental visual cues and characteristics of plants found in agricultural facilities for the purpose of rapid automated identification of plant species. The human eye, coupled with the brain?s processing power , can readily distinguish between different plant species. This capability was one of the basic needs for humans to become an agrarian society (farming requires weeding), which helped start enormous social advancement. Similarly, to bring automated systems to the next generation of capability, computer vision must interact with the natural world with greater fidelity. Today?s computer vision has ability to detect a ?splotch? of vegetation versus no vegetation. This project will advance computer vision by developing the equipment and software algorithms necessary to automatically distinguish plant types. The project team will build a computer vision algorithm based on a field customized support vector machine (SVM) that can automatically and reliably identify a known crop versus a foreign plant (i.e. weed) for use in a larger system for automated weeding. By creating the ability for computers to distinguish between plant types, we will enable food to be grown with reduced amounts of chemical herbicides. The broader impact/ commercial potential of this project is to increase the competitiveness of vegetable farms, particularly organic ones, while improving human health and the environment. Today, organic farms represent 5% of the U.S. agricultural economy and are growing at a pace to double organic acreage every 4 years. A key feature of organic farming is the lack of herbicides. Consequently, organic farms are normally weeded by hand. Weed control represents approximately 50% of operating costs for organic farms, compared to less than 10% for conventional ones. With an estimated $700M spent annually on weeding organic farms, there is a substantial commercial opportunity to create a system that can weed farms automatically. This project will develop a system that uses a computer system towed behind a tractor to automatically detect and eliminate weeds at early plant stages. The system can be developed and deployed at less than 1/5 the life-cycle costs of hand weeding. The technology is also applicable to conventional crop thinning where it can significantly reduce the amount of herbicides used. Additionally this technology has a profound health and sustainability benefits by eliminating human exposure to chemical herbicides through food and avoids herbicides leaching into the soil.