We present a comprehensive study of LED-based optical sensing systems for the classification of waste materials, analyzing recent developments in the field. Accurate identification of materials such as plastics, glass, aluminum, and paper is a crucial yet challenging task in waste management for recycling. The first approach uses short-wave infrared reflectance spectroscopy with commercial Germanium photodetectors and selected LEDs to keep data complexity and cost at a minimum while achieving classification accuracies up to 98% with machine learning algorithms. The second system employes a voltage-tunable Germanium-on-Silicon photodetector that operates across a broader spectral range (400–1600 nm), in combination with three LEDs in both the visible and short-wave infrared bands. This configuration enables an adaptive spectral response and simplifies the optical setup, supporting energy-efficient and scalable integration. Accuracies up to 99% were obtained with the aid of machine learning algorithms. Across all systems, the strategic use of low-cost LEDs as light sources and compact optical sensors demonstrates the potential of light-emitting devices in the implementation of compact, intelligent, and sustainable solutions for real-time material recognition. This article explores the design, characterization, and performance of such systems, providing insights into the way light-emitting and optoelectronic components can be leveraged for advanced sensing in waste classification applications.

Reliable fire detection is pivotal for safety in both domestic and industrial environments. Traditional detection methods often suffer from false alarms and environmental interference, thus proving inadequate in critical scenarios. In this study we propose a novel flame detector based on a voltage-tunable dual-band Ge-on-Si photodetector combining a wavelength sensitive detection with classification algorithms. We evaluated its performance against various interference sources, exploiting machine learning models such as Support Vector Machines (SVM) and Quadratic SVM, achieving up to 99.8% classification accuracy. We demonstrate the system can effectively distinguish flames from background illumination. In addition, the sensor is able to detect hot objects, making the device suitable for early hazard detection, as well. This combination of the unique optical property of the wavelength sensitive photodetector and powerful machine learning algorithms may represent a significant advancement in fire detection technology.