Leaf Wetness Sensor Data Processing: From Raw Signal to Accurate Classification

We have a leaf wetness sensor that measures leaf wetness by detecting changes in the dielectric constant of its surface. The sensor's analog output provides a continuous signal representing the degree of wetness, while the digital temperature output compensates for environmental temperature variations. The high signal-to-noise ratio (SNR) of the sensor ensures precise measurements, even in challenging conditions like fog or light dew. 



Challenges with Raw Sensor Data

The raw analog sensor signal is susceptible to noise and temperature variations, making it difficult to accurately correlate sensor readings with specific leaf wetness events like rainfall or dew. Without proper processing, the time-series data can be misleading, as it may be difficult to distinguish between true moisture signals and background noise. 


Data Processing Pipeline

To overcome the limitations of raw sensor data, I developed a data processing pipeline that involves several key steps: temperature compensation to account for temperature-induced variations, noise reduction to eliminate unwanted signal fluctuations, and baseline correction to establish a reference point for dry conditions. By applying these techniques, the processed data provides a clear representation of leaf wetness events, making it easier to identify and analyze. 

Here's a breakdown of the leaf wetness sensor data analysis pipeline:

 𝟭. 𝗦𝗶𝗴𝗻𝗮𝗹 𝗣𝗿𝗲-𝗽𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴: This involves initial steps like analog-to-digital conversion, temperature compensation, and noise reduction. The goal is to clean and prepare the raw sensor data for further analysis. 

 𝟮. 𝗦𝗶𝗴𝗻𝗮𝗹 𝗣𝗿𝗼𝗰𝗲𝘀𝘀𝗶𝗻𝗴: This includes steps like baseline calculation, feature extraction, and normalization. These steps transform the raw data into a suitable format for machine learning algorithms. 

 𝟯. 𝗦𝗶𝗴𝗻𝗮𝗹 𝗖𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻: The processed data is fed into a machine learning model, which classifies the signal into different categories (e.g., rainfall, dew, frost). 


Model Development and Results

Through extensive data collection and analysis, I developed a piecewise linear model to compensate for temperature-induced variations in sensor readings. You can learn more about the effect of temperature on leaf wetness sensor readings here.

Additionally, I trained a machine learning model to accurately classify leaf wetness events into categories such as rainfall, dew, and frost. The model, which leverages leaf wetness, sensor temperature, and time as features, achieved an impressive 94% accuracy. While adding soil moisture and temperature improved accuracy to 99%, I am focused on optimizing the model's performance using only leaf wetness sensor data. 

I plan to share more details in a future article.

Comments

Post a Comment

Popular posts from this blog

Leveraging Microclimate Data, Proximal Canopy Sensing, and Machine Learning for Precise Soil Water Potential Prediction

Image
  Introduction Accurately predicting soil water potential is crucial for optimizing irrigation practices and minimizing water waste in agriculture. This project delves into the critical relationship between high resolution, in-field microclimate measurements, canopy surface temperature , time of day, and soil water content and potential. Building upon our previous work in biophysical modeling , we employ advanced machine learning techniques to investigate the impact of measurement timing on prediction accuracy. Enhanced Prediction with Machine Learning This project utilizes a range of machine learning algorithms, including: Linear Regression:  Establishes a linear relationship between features and target variables. Decision Tree Regressor:  Makes decisions based on a series of if-else questions. Random Forest Regressor:  An ensemble method combining multiple decision trees for improved accuracy and reduced overfitting. Gradient Boosting Regressor:  Iteratively b...
Read more

The Rise of the "Human Data Farms": A Satirical Look at the Future of AI

Image
  The race for Artificial General Intelligence (AGI) has reached a fever pitch, with tech giants vying for dominance in the AI landscape. As Elon Musk has pointed out ( https://www.theguardian.com/technology/2025/jan/09/elon-musk-data-ai-training-artificial-intelligence ), the increasing scarcity of data for training AI models is a significant challenge. But as data becomes increasingly scarce, a new and perhaps more sinister solution is emerging: the rise of the "Data Farms." Imagine a dystopian future where tech companies, desperate for the raw materials needed to fuel their AI models, begin constructing self-contained communities. These "Human Data Farms" would resemble a hyper-surveilled version of The Truman Show, where residents are paid handsomely to live their lives under constant observation. Every interaction, every conversation, every emotion – all meticulously recorded and fed into the insatiable maw of AI algorithms. This scenario, while seemingly absur...
Read more

For CEA Growers: Tips for Substrate Monitoring and Automation

Image
  Based on my communications and meetings with our customers on solutions for substrate monitoring and decision making in controlled environment agriculture (CEA), I have identified a numbers of important areas to briefly discuss:   1) Sensor quantity optimization Most growers do not need a lot of sensors to monitor their grow rooms. There are good ways of determining the optimum number of sensors required for different applications. In addition to an unnecessarily high cost, a large number of sensors in one facility can create serious design, installation and maintenance headaches for different parties involved. More data does not translate into better insight or decision-making unless we know how to deal with “big data”. Large amount of data that grower does not know what to do with leads to frustration and mistrust in the technology.   More data does not translate into better insight or decisions unless we know how to deal with “big data”. Large amount of data that gro...
Read more