A minirhizotron is an in situ root imaging system that captures components of root system architecture dynamics over time. Commercial minirhizotrons are expensive, limited to white-light imaging, and often need human intervention. The implementation of a minirhizotron needs to be low cost, automated, and customizable to be effective and widely adopted. We present a newly designed root imaging system called SoilCam that addresses the above mentioned limitations. The imaging system is multi-modal, i.e., it supports both conventional white-light and multispectral imaging, with fully automated operations for long-term in-situ monitoring using wireless control and access. The system is capable of taking 360° images covering the entire area surrounding the tube. The image sensor can be customized depending on the spectral imaging requirements. The maximum achievable image quality of the system is 8 MP (Mega Pixel)/picture, which is equivalent to a 2500 DPI (dots per inch) image resolution. The length of time in the field can be extended with a rechargeable battery and solar panel connectivity. Offline image-processing software, with several image enhancement algorithms to eliminate motion blur and geometric distortion and to reconstruct the 360° panoramic view, is also presented. The system is tested in the field by imaging canola roots to show the performance advantages over commercial systems.

Chlorophyll-A concentration is one of the most commonly measured water quality parameters. It is an indicator of algal biomass and provides insight into stressors such as eutrophication and bloom risk. It is also a widely used metric in terrestrial ecosystems as an indicator of photosynthetic activity and nutrient limitation. Laboratory-based methods for measuring chlorophyll-A require expensive instrumentation. In this paper, we proposed a smart, low-cost, and portable smart sensor system to measure the concentration of chlorophyll-A in an extracted solution using two consumer-grade spectral sensors that read the reflectance at 12 discrete wavelengths in visible and near-infrared spectra. The system was tuned for an optimal distance from the sensors to the solution and an enclosure was printed to maintain the distance, as well as to avoid natural light interference. Extracted chlorophyll solutions of 51 different concentrations were prepared, and at least 100 readings per sample were taken using our smart sensor system. The ground truth values of the samples were measured in the laboratory using Thermo Nano 2000C. After cleaning the anomalous data, different machine learning models were trained to determine the significant wavelengths that contribute most towards chlorophyll-A measurement. Finally, a decision tree model with 5 important features was chosen based on the lowest Root Mean Square and Mean Absolute Error when it was tested on the validation set. Our final model resulted in a mean error of 10.9 lg/L when applied on our test set. The total cost was around $150.