The plant health monitor tool is intended to present plant health information correlated with geometric location information after a particular monitoring run at Rivendale farms. In order to understand the required features for this tool, the first step was to list down the customer benefits associated with this tool. The identified benefits to the customer are as follows:
1) View relevant plant health information efficiently
2) Give user-driven feedback to provide expert supervision to the deep learning based detections.
The second step was to identify the high-level features which are required from this tool which can be summarized as follows:
1) 2D grid representation of plant health data – Display plant health information collected correlated to its geometric location. This can be done be representing healthy plants with one color and unhealthy plants with another color.
2) User in the loop – Allow the user (aka the farmer) to modify the disease level for a particular pixel. This is done in order to remove false positives/ outliers which may arise due to edge cases with the deep learning pipeline.
Subsystem breakdown
The next step involved breaking down the high-level features of the system into sub-systems. The main subsystems identified from this process are as follows:
1) ROS Bag Parser– This subsystem takes a ROS Bag as input. This ROS Bag consists of location data and image data associated with their related timestamps. The location data was obtained using an RTK GPS in one of our previous runs. The image data consists of the right and left images from the stereo camera. The data association step looks at sensor data published at the same timestamp and converts them into a dictionary mapping.
2) Mask R-CNN Pipeline– The Mask- RCNN pipeline outputs labels from the deep network pipeline will be taken as input to the map visualizer.
3) Matplotlib Visualizer– The Matplotlib visualizer blocks takes the correlated geometric and image information and convert it into a visual representation.
