Abstract:
Pastures are an indispensable part of agriculture which contributes to
industries of meat and milk. Among the problems affecting dairy farmers are
weeds. Weeds constraint the resources and space for grass growth in pastures.
The traditional methods of using chemical fertilizers and pesticides to control
weeds have various limitations. Precise control of weeds can help dairy farmers
grow better. Processing sensor data of pastures and monitoring them with image
processing can help in better control with less human error.
This study is reviews the pasture weed detection and technologies for deploying
weed models on the Internet such as docker, mage, and microservice. The two
architectures mentioned in this study are server-based and client-based. The
server-based architecture means the farmer will collect weed images and upload
them to the cloud server. The cloud will complete all processes and return the
results to the client. In the client-based architecture, most of the data processing
is performed on the client-side, while the essential data are transferred to the
server to transfer to other architecture components.
This study analysed two federated learning models and their attributes and
benefits to our weed management systems. We also thought of a client-based
model, which means that farmers use mobile devices to detect weeds through the
weed model. In addition, for the client-server model, farmers use mobile devices
to upload weed images to the server and get feedback. Their interconnections can
help for a better understanding of our analysis.
The weed system deployment will help the farmers in better weed management
while using technologies that can support a better accuracy in weed clean-up. Also,
the deployment will help dairy farmers to automate their weed monitoring tasks
and collaborate and share their weed management experiences with other
farmers.