Development of a low-cost portable device for pixel-wise leaf SPAD estimation and blade-level SPAD distribution visualization using color sensing

Soil Plant Analysis Development (SPAD) was proved as a reliable indicator for the representation of leaf chlorophyll or nitrogen content. Portable plant SPAD sensors have been widely studied to determine plant nitrogen status rapidly in recent years. However, the SPAD-502 device could only measure a single point in one shot, and its price was high. Moreover, most of the presented DIY devices for SPAD evaluation required clamping operation to get the transmitted light signal using complex hardware architecture. In this study, an ultra-portable SPAD evaluation system, defined as SPAD-Cap, was proposed for leaf SPAD distribution analysis. The RGB camera module, Raspberry Pi Zero, and light source were integrated. Two modeling methods, including partial least square regression (PLSR) and convolutional neural network for regression (CNN-R), were used to evaluate pixel level SPAD value based on the color features of the corresponding pixel. The SPAD distribution map was generated by traversing all pixels of the tested blade sample using the trained model. A Web GUI was developed for controlling the whole system and accessing the visualization result. The testing result showed that the designed SPAD-Cap achieved R-2 of 0.97 and RMSE of 2.5 SPAD units to inspect rape leaf SPAD. For leaves in several species, R-2 was 0.91, and RMSE was about 4.0 SPAD. Moreover, all the processing steps, including data collection, processing, storage, and visualization, were involved in the SPAD-Cap system, which could be a potential solution for rapid and automatic inspection of leaf SPAD distribution.

Automated summary

A ultra-portable SPAD evaluation system, SPAD-Cap, was proposed for leaf SPAD distribution analysis using an integrated RGB camera module, Raspberry Pi Zero, and light source. Two modeling methods, PLSR and CNN-R, were used to evaluate pixel level SPAD value based on color features. The SPAD distribution map was generated by traversing all pixels of the tested blade sample using the trained model.