Technology is fundamental, and innovation is the soul. The “Super Base Team” will participate in the scientific fund project in 2024, aiming to build a low-cost high-throughput phenotyping research system based on Raspberry Pi computers and cameras. This automatic imaging analysis system for plant seedlings provides an efficient and economical solution for large-scale plant phenotypic analysis, greatly reducing the consumption of human resources.
01
Preliminary Research
During the preliminary research process, the team ensured that the processing target of the automatic imaging analysis system—morphological features—remains insightful for future related research. Team members first had an offline discussion with the advisor Zhu Ziqiang and found relevant literature based on their phenotypic research. The team members have now drafted and summarized the morphological features and are conducting in-depth learning for the next steps regarding the acquisition and analysis of related morphological features.
During the winter vacation, the team members collected relevant literature in the field, determined the design ideas for the imaging analysis system, planned to use 3D-printed parts as structural components, and employed low-cost Raspberry Pi development boards and cameras to achieve automated imaging through remote protocols such as VNC. Finally, they intend to develop based on third-party libraries such as OpenCV and scikit-image and build convolutional neural networks to achieve automatic image analysis processing.
Swipe left or right to see more images
02
Preliminary Experimental Work
In the hardware implementation part, the team purchased the relevant materials to assemble the system in March. Team members are learning knowledge about operating the Raspberry Pi development board with a computer and using Python to operate the Raspberry Pi camera to better assemble the device.
Team members are learning to operate the Raspberry Pi camera
Team members planting rapeseed
Preliminary photos taken with the Raspberry Pi camera
Team members learning to use the Raspberry Pi development board
Swipe left or right to see more
In the software implementation part, to achieve the function of image conversion and analysis of various representations, team members learned the development and use of third-party libraries such as OpenCV and scikit-image during the winter vacation, and are currently studying PyTorch and TensorFlow for building convolutional neural networks.
Preliminary training of the convolutional neural network layer diagram obtained, used for edge detection of Arabidopsis seedling images.
The deviation from the samples gradually decreases during the training process, but due to the limited sample size, the error rate remains high.
Based on the above model, preliminary edge inspection of the seedlings was implemented, which still needs optimization.
The team members stated: “We hope to promote innovation in the field of biology through this low-cost high-throughput phenotyping research method, and we also hope that this method will provide researchers with a more convenient and economical approach.”
We hope that through the team’s efforts and innovations, this low-cost high-throughput phenotyping research system can be realized and put into use, providing more convenient and economical tools and resources to solve important problems in biology, contributing our part.
Image | Kong Weiji, Wan Zhongxiang, Tang Yan, Ma Feixiang
Text | Tang Yan
Editor | Tang Yan
Review | Mu Xu, Zhang Yuchen