Accurate Imputation of Greenhouse Environment Data for Data Integrity Utilizing Two-Dimensional Convolutional Neural...

Area of Science:

• Agricultural Engineering
• Data Science
• Environmental Monitoring

Background:

• Greenhouse microclimate monitoring is crucial for resource efficiency but faces challenges with sensor reliability in harsh environments.
• Accurate environmental data is essential for optimizing crop productivity and quality in controlled agricultural settings.

Purpose of the Study:

• To evaluate the efficacy of a U-Net convolutional neural network (ConvNet) architecture for imputing missing tabular environmental data from greenhouses.
• To assess the performance of the U-Net model under various data-loss scenarios, including individual and complete sensor failures.

Main Methods:

• Utilized a U-Net ConvNet architecture for data imputation on tabular datasets collected from greenhouse environments.
• Simulated different data-loss conditions, mimicking sensor malfunctions and complete data absence.
• Trained and tested the U-Net model with a specific screen size (U-Net50) to analyze its imputation accuracy.

Main Results:

• The U-Net50 model demonstrated superior performance, achieving the highest coefficient of determination and lowest root-mean-square errors across all environmental factors analyzed.
• The U-Net architecture successfully learned and replicated the dynamic patterns of the greenhouse microclimate from the training data.
• The model proved effective in imputing data even under conditions of complete sensor failure.

Conclusions:

• The U-Net architecture is a viable and effective tool for imputing missing tabular data in greenhouse environmental monitoring systems.
• Successful data imputation using U-Net can enhance data integrity, potentially leading to increased crop productivity and improved quality in greenhouses.
• Proper training of the U-Net model is critical for accurate imputation of greenhouse environmental data.