Multimodal Data Fusion for Early Stress Detection in Indoor Plants: a data-driven approach for monitoring plant health

Read full text Summarise with VIRAT
Authors: Sachini Nimesha Fernando
Cited by: 0
References: 22

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Indoor plants often experience stress due to the limited and controlled environmental conditions. Recognizing early stress was significant for maintaining plant health and improving care routines. The goal was to investigate whether a low-cost multimodal dataset could detect the early signs of indoor plant stress before human observation. A controlled study was conducted with five indoor plant species grown in two types of soil mixtures under normal and stress watering conditions over a period of 70 days. Environmental features such as soil moisture, humidity, light intensity, temperature, and water level were measured using a manual meter, a smartphone application, and an affordable sensor, while daily image data were analyzed to compute leaf greenness. Data preprocessing was conducted by linking the image data to the sensor data, computing leaf greenness through image feature extraction, cleaning the data by handling missing values and outliers, and finally, normalizing the data using the min-max method. Exploratory data analysis discovered temporal trends and a weak correlation between environmental features and plant stress, demonstrating that stress was influenced by a combination of factors. Statistical analysis confirmed that soil type significantly affected the soil moisture, which influenced the plant stress directly across the species. Two models were created to predict the stress levels. Human-based model achieved moderate accuracy (44-91%), with limitations in identifying mild stress due to an imbalanced dataset. Sensor-based model achieved high accuracy (98-100%), demonstrating the ability to detect the early stress before the visual observation. Feature importance analysis confirmed the three most influential factors that affect the plant stress, such as leaf greenness, humidity, and soil moisture. Finally, the above results demonstrated that a multimodal sensor-based approach can provide an earlier and more reliable detection of indoor plant stress compared to visual observation, supporting small indoor gardening, smart buildings, and automated plant monitoring systems.

Fields of Study & Subjects

Smart Agriculture and AI Plant and Biological Electrophysiology Studies Leaf Properties and Growth Measurement

Publication details

Journal / source
Theseus (Ammattikorkeakoulujen)
Publication date
1 January 2026
Language
EN
Document type
Thesis