機器學習應用於光電感測元件之研究

Abstract

機器學習技術與光電元件的結合已成為提升現代感測平台功能性、智慧性和適用性的變革性方法。受低功耗、高選擇性和即時感測需求的驅動，本論文對機器學習在室溫條件下運行的光電感測器中的應用進行了全面的研究。 探討了三種相互關聯的感測策略。首先開發了一種光化學激活氣體感測平台，利用紫外線激發實現穩定、高效的氣體檢測，而無需加熱。其次，提出了一種光學調製氣體識別系統，其中透過改變照明強度產生的不同光學指紋透過機器學習模型進行分析，包括支援向量機、樸素貝葉斯、隨機森林和 K 近鄰演算法，以實現準確的多氣體分類和濃度預測。第三，製造了基於自供電異質接面結構的寬頻光電探測器陣列，展示了使用機器學習輔助波長識別對紫外線、可見光和近紅外線區域的多光譜光訊號進行分類和重建的能力。 透過將材料創新與機器學習驅動的數據處理相結合，這項工作建立了智慧氣體感測和光學檢測的統一框架，促進了可擴展、節能和高度適應的下一代感測技術。研究結果強調了機器學習在提高光電設備性能和多功能性方面的關鍵作用，為未來的智慧感測應用奠定了基礎。

The integration of machine learning (ML) techniques with optoelectronic devices is emerging as a transformative approach to advancing the functionality, intelligence, and applicability of modern sensing platforms. Motivated by the demand for low-power, high-selectivity, and real-time sensing, this dissertation presents a comprehensive study on the application of ML in optoelectronic sensors operating under room temperature (RT) conditions. Three interrelated sensing strategies are explored. First, a photochemically activated gas sensing platform is developed, utilizing ultraviolet (UV) excitation to enable stable and efficient gas detection without the need for thermal heating. Second, an optically modulated gas identification system is proposed, in which distinct optical fingerprints generated through varying illumination intensities are analyzed by ML including Support Vector Machines (SVM), Random Forest (RF), Naïve Bayes (NB), and K-Nearest Neighbors (KNN) to achieve accurate multi-gas classification and concentration prediction. Third, a broadband photodetector array (PDA) based on a self-powered heterojunction structure is fabricated, demonstrating the ability to classify and reconstruct multispectral light signals across UV, visible (Vis), and near-infrared (NIR) regions using ML-assisted wavelength recognition. By integrating material innovation with ML-driven data processing, this work establishes a unified framework for intelligent gas sensing and optical detection, promoting scalable, energy-efficient, and highly adaptive next-generation sensing technologies. The results underline the critical role of ML in enhancing the performance and versatility of optoelectronic devices, laying a foundation for future smart sensing applications.