|
Precision Agriculture stands out as one of the most promising areas for the development of new technologies around the world. Some advances from this area include the mapping of productivity areas and the development of sensors for climate and soil analysis, improving the smart use of resources during crop management and helping farmers during the decision-making stages. Among the problems of modern agriculture, the intensive and non-localized use of herbicides causes environmental issues, contributes to elevated costs in farmers’ budgets and results in applications of chemical substances in non-target organisms. Although there are many selective herbicide spraying systems available for use, the majority working principle is based upon chlorophyll detectors, thus not being able to distinguish crop plants from weeds with high accuracy in crop’s post-emergence herbicide applications (“green-on-green” application). The main objective of this study is to develop a multispectral camera system for in-crop weed recognition using Computer Vision techniques. The system was built with four monochromatic CMOS sensor cameras with monochromatic wavelength bandpass filters (green, red, near infrared and infrared) and a RGB camera. Soybean and weed plants images were captured in a controlled environment using an automated v-slot rail system to simulate the movement of a spray tractor in the field. Infrared images presented higher precision (90.5%) and recall (89.3%) values compared to the other monochromatic bands, followed by RGB (87.0% and 86.1%, respectively) and near infrared images (83.6% and 87.9%), suggesting that infrared wavelengths plays an important role in plant detection and classification. Our results state that the combination of Computer Vision and multispectral images of plants is a more efficient approach for targeting weeds among crop plants for post-emergence herbicide applications.
|
