Image processing and machine learning for data investigation in flash memory

Abstract

As the world's data continues to grow, memory storage is essential for retaining and accessing information. Flash memory, a type of non-volatile storage, is used for permanent data storage. However, data security is of great concern due to the potential for data loss through deletion, encryption, or corruption. Digital forensics is an important tool for analyzing and preserving digital evidence in the event of a cyberattack or data breach. This project focuses on investigating data stored in flash memory of unfunctional electronic devices using image processing and machine learning techniques. The report includes a literature review of image processing, machine learning, as well as a detailed methodology on segmentation using U-Net, designing a localization algorithm, and using LeNet-5 for classification of data in flash memory into binaries 1’s and 0’s. Experimental results are presented and discussed, and recommendations for future work are provided.

The use of U-Net model and LeNet-5 model in segmentation and classification of the images resulted in high accuracy of 98.32% and 99.50%, respectively. The algorithms developed in this study have significant implications for digital forensics investigations, particularly in cases where digital evidence is compromised due to deletion, modification, or corruption. The use of image processing and machine learning techniques can provide valuable information for identifying the causes and possible intent of an attack, preventing data losses, and preserving digital evidence to solve technology-related crimes.