AI vs. Trash: The Effect of Luminosity on the Accuracy of an AI Model

Overview

AI vs. Trash: The Effect of Lighting on Accuracy of an AI Classification Model Louis Lu This study investigated the effect of lighting intensity on the classification accuracy of a fully trained artificial intelligence (AI) model designed to sort waste materials. It was hypothesized that higher luminosity levels would result in greater classification accuracy. This hypothesis was grounded in established optical and computational principles, including luminosity, illuminance (lux), point spread function, glare effects, and the performance characteristics of the five algorithms integrated within the AI model. To test this hypothesis, a mechanical sorting system was constructed. Two conveyor belt tracks were designed and 3D-printed, assembled into a single continuous belt, and integrated with a motor driver, microcontroller, and stepper motor. An AI classification model was developed with code assistance from Claude and implemented using a Gemini API key. A Raspberry Pi controlled a servo motor attached to a sorting arm, enabling automated physical separation of classified waste items. The experiment consisted of five trials conducted under four different lighting conditions, ranging from 500 to 2000 lumens. During each trial, waste items were placed on the conveyor belt beneath a camera, and the AI classified and sorted the materials. Results supported the hypothesis. The highest lighting condition (2000 lumens) achieved 100% classification accuracy across all trials. The 1000-lumen condition produced an average accuracy of 96%, while the 500-lumen condition yielded an average accuracy of 86.67%. These findings indicate that increased lighting intensity significantly improves AI classification performance in waste-sorting applications.

Competition history

• CSEF 2026 Environmental Engineering (Junior Division) · Entry J-11-09

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Environmental Engineering projects
• CSEF Environmental Engineering projects
• Every CSEF 2026 project