The Brazilian Agricultural Research Corporation Embrapa has developed a solution that uses drones for detecting and counting plants.
The method combines computing vision and deep learning. Trials have been done in corn and citrus plantations. In the case of corn, 33,360 plants in 224 crop rows were identified counted. According to Embrapa, evaluation showed 94% accuracy. In citrus 97 of 100 trees were correctly identified.
In both cases, Embrapa says their system provides better accuracy than visual counting and even existing software solutions. The International Society for Photogrametry and Remote Sensoring (ISPRS) published these results in the Journal of Photogrammetry and Remote Sensing.
The method replaces visual inspections, which are slow, laborious and biased, says Embrapa. Another advantage in relation to traditional practices is that countings can be done in each planted area in its entirety. This makes it easier for growers to manage their fields and optimise fertilising or crop protection operations. Additionally, it also makes harvest forecasts possible.
It’s all being made possible with a convolutional neural network (CNN) for counting plants and crop rows simultaneously through images made by drones, that are equipped with RGB cameras.
Text continues underneath image
The RGB system is a low-cost solution because most drones are equipped with it as standard. “RGB usage reduces costs when compared to sensors in other ranges of the light spectrum. Thus, the method is a low-cost and viable alternative to be applied to any crop”, said Lucio Jorge, researcher at Embrapa. According to him the great advantage of the nely developed system is the software, which allows for processing of the data in real-time.
The project involved researchers from 3 Brazilian universities (Federal of Mato Grosso, Oeste Paulista and Santa Catarina State), the Canadian Waterloo University and specialists from Embrapa Instrumentação (Instrumentation).