CNN-Assisted Growth Monitoring and Stress Management of Cucumber in Semi-Transparent PV Greenhouses for Agrivoltaics

Currently operating commercial photovoltaics (PV) systems integrated with agricultural production (Agrivoltaics) offer immense potential for the dual harvest of renewable energy and agro-products. Within controlled-environment agriculture (CEA), the use of semi-transparent photovoltaics (ST-PV) and the ability to control the microclimate and shading are beneficial for the production of high-value crops such as cucumbers. The objective of this research was to commence the cultivation of cucumbers under evolving CEA-PV systems by combining greenhouse experiments with computer vision (CV) based driven phenotyping to create an analytical framework and system control framework for the cultivation of cucumbers in an evolving CEA-PV system. The method involved using the monitored plant vigor to control in real time the irrigation and shading of the cucumber plants. The control of irrigation and shading was based on the monitored plant vigor as determined by a U-Net++ implementation for canopy segmentation, an EfficientNet-B3 implementation for stress detection, and a CNN regressor for growth trait estimation. Within the greenhouse, uniform environmental and fertigation conditions were established to evaluate the effect of four shading regimes (0%, 20%, 40%, 60%) on the cucumbers. Simulated, yet representative results predicted cucumber yields to be stable (within ±4% of full yield) with a 20% shading and a 15-20% reduction in water use compared to full sun. Yield was also observed to drop by 10-14% under higher shading of 40 to 60% due to insufficient photosynthetic activity for fruiting. The CNN based models were robust, (segmentation IoU 0.91, stress-class F1 0.92, LAI regression R²≈0.93), allowing for precise and comprehensive monitoring in an annual non-invasive fashion. The greenhouse's annual photovoltaic (PV) output was estimated to be 1,550 to 1,750 kWh/kWp which is able to exceed the energy demand resulting to a net energy surplus. The outcome demonstrates that the cucumber crop can be successfully combined with controlled environment agrovoltaic systems with moderate shading for optimum cucumber yield. Moreover, informed supervision through Artificial Intelligence (AI) helps to navigate closed-loop systems and enhance the water-use efficiency and yield stability.