Development of Image Classification Model for Product Package Recognition in Industrial Environments

High Level Description

The goal of this project is to develop a system for identifying and classifying objects on a conveyor belt into specific Stock Keeping Units (SKUs). The objects are consumer product packaging sourced from household waste, and the aim is to analyze the frequency of specific items consumed by households—valuable data for production companies. Additionally, the system will assist in sorting waste based on packaging material.

This thesis focuses on the development and evaluation of a classification model to categorize the objects into product types and corresponding SKUs. A key part of the project will involve comparing inference methods across different datasets.

Project Description 

The setup includes an industrial-grade camera positioned above a moving conveyor belt. The camera captures images when an object is detected passing through. These images must be classified into their respective SKUs or material categories. Alongside this, a set of high-definition images of the objects, captured by a handheld camera, is manually labelled with SKU details and product information.

Given the availability of a large dataset of labelled high definition (HD) images, the objective is to implement a classification model trained on these images, which can then be used to classify the lower-quality images from the industrial camera. The core aim is to develop and test an inference model to accomplish this task. Further, the model will be optimized for speed and efficiency. 

The project objectives are:

• Classify packaging material for waste sorting.

• Identify product brand, type, and other details.

• Recognize and assign the correct SKU to each item.

For the inference model, the study will explore and implement techniques such as domain adaptation, domain adversarial training, and various data augmentation methods, aiming to enhance the model’s performance if improvements are anticipated.

Who are we looking for?

We are looking for candidates with a strong background in computer vision and machine learning, specifically in image classification and deep learning. This thesis is ideal for students pursuing degrees in Computer Science, Artificial Intelligence, or related fields with a focus on Data Science and Machine Learning.

Knowledge of convolutional neural networks (CNNs), data preprocessing, and experience with Python-based frameworks like TensorFlow or PyTorch will be essential. Familiarity with domain adaptation, and data augmentation techniques would be an advantage. Additional experience with hardware integration for vision systems, such as cameras and data acquisition, is a plus.

Purpose

The purpose of this thesis is to develop an efficient image classification system to recognize and categorize consumer product packaging on a moving conveyor belt. This system will identify products by their SKUs and packaging material, providing valuable data for product usage analysis and waste sorting. The model will be designed for industrial applications, focusing on achieving high accuracy in diverse conditions and improving the speed and efficiency of inference in real-time environments.

The thesis project can be published and used in your personal portfolio as well as in company marketing. Include Resumé/CV and portfolio in your application.