Denoising Autoencoders (DAEs) are a machine learning tool used to reconstruct noisy data in a way that reduces noise, allowing artificial intelligence systems to better recognize patterns in data or images. An autoencoder model consists of symmetrically arranged layers, which contain different numbers of "nodes" (processing units) and sequentially process the data. This research demonstrates the effectiveness of a DAE for a context-aware robotic hand that uses a near-infrared sensor to distinguish between five material types (wood, ceramic, cardboard, plastic, and stainless steel). The accuracy of the robot's material recognition is greatly affected by ambient light, distance between the sensor and the object, and the orientation of the sensor relative to the object. Current programming methods don't resolve this problem. In this research, near-infrared material scans will be used to demonstrate the need for a denoising autoencoder for the robot. Multiple autoencoder models, containing different numbers of nodes per layer, will be tested on near-infrared scan data to evaluate their accuracy, and thus improve the accuracy of the material recognition.