Research need from Focus Group: Individual plant recognition
non-chemical weed control
Plant production and horticulture
This is the problem:
This need for research has been identified in FG 32 on “Non-chemical weed management”. If you want to know more about it, check the final report of the Focus Group.
Individual plant recognition for more precision would be a prerequisite for site-specific weed management. For the purpose of single plant detection and non-chemical weed control several aspects need to be distinguished. When control of weeds is required with minimal inputs of energy, and in a sustainable way not disturbing more soil than necessary, only individual weed plant recognition is a good solution. Detection of crop plants is not enough for effective and efficient weed control with minimal inputs of energy, as large bare soil areas will be treated or disturbed, having negative impacts on the growth conditions of the crop. One can think of excessive evaporation, loss of water or germination of new weed seeds due to soil disturbance effects. For the purpose of single weed plant identification and localisation, several techniques are available. As of 2019 large databases are available containing images of weed species in several growth stages. These databases can be used to train real-time deep learning neural networks. For training deep learning classification networks, large numbers of weed example images are necessary, and adaptation of training data is necessary to handle new crop and weed types during the season and between seasons.