Vision statement: Promote the development and use of sensor-based weed recognition capability for advancing weed research and precision weed control applications

Technological advancements in plant imaging and artificial intelligence/machine learning (AI/ML) have enabled the capability for accurate recognition of individual weed plants and plant parts as they occur within agricultural contexts. This capability facilitates opportunities for efficient weed management through the precision targeting of individual plants and plant parts with an expanded range of control technologies including herbicides and non-chemical tools. Further, these technologies can advance our understanding of weed ecology, through the potential for highly detailed and efficient examination of agricultural environments and weed-crop interactions. The development of weed recognition technologies will have an immense impact on the field of weed science, with the potential to completely transform weed research and the ability to control weeds.

Weed recognition capability using an ML-based approach occurs in three key steps commencing with image collection, followed by image classification/annotation, and then algorithm training on the image dataset. Weed recognition algorithm development is, therefore, founded as well as completely reliant on the availability of suitable weed image datasets. The image dataset requirements for algorithm training increase with the complexity of the scenarios in which weeds occur. Simple scenarios may require just a few thousand images, while more complex scenarios (e.g., multiple grass seedlings in a mix) will require a substantially larger number of suitably annotated images. Despite the exciting weed research and control opportunities that come with digital technologies, the limited availability of suitable image datasets and recognition algorithms remains a major constraint. Moreover, currently available datasets and algorithms are scattered and often less structured, making them difficult to access in a centralized, searchable, and user-friendly format. Additionally, providing adequate training and outreach is critical for promoting the development and broad application of these technologies.