A Hybrid Approach Combining Neural Networks and Signal Processing for Indoor Temperature Detection Using Plants’ Bioelectric Potential

Indoor temperature detection is traditionally performed using standard sensors and monitoring devices, which can be cost-prohibitive and inefficient for large-scale or long-term applications. This study proposes a novel approach to ambient temperature detection utilizing plant bioelectric potential signals. This study presents a hybrid model integrating FFT, CNN, and LSTM to classify temperature conditions based on plant bioelectric signals. FFT is employed to enhance signal clarity by filtering noise and extracting frequency components. CNN captures spatial characteristics of the data, while LSTM models temporal dependencies in sequential signals. The model processes data in segments with varying window sizes (16, 32, 64, 128, 256) and overlap levels (0 %, 25 %, 50 %, 75 %) to ensure flexibility and robustness. Experimental results demonstrate that the hybrid model outperforms traditional approaches, achieving classification accuracy of up to 100%. This highlights the potential of plant bioelectric signals as a sustainable and cost-effective alternative to conventional temperature sensors. Additionally, the methodology shows promise for analyzing other one-dimensional signals, paving the way for advanced applications in environmental monitoring and precision agriculture.