Machine learning optimizes information gathering in logistics processes

A large industrial partner (international logistics service provider) was faced with the challenge of efficiently realizing the classification of logistics processes, especially in the packaging of parcels. The aim of the collaboration with the Software & Information Engineering department was to develop a classifier for quantity forecasts.

Use of machine learning for automation

In contrast to traditional programming methods, where rules have to be created manually, machine learning made it possible to find rules automatically by training them with real data. The aim was for the system to learn rules independently in order to classify the visual properties of parcels and improve quantity forecasts.

Advantages of the machine learning approach

• Reduced development effort compared to traditional software programming
• Automatic adaptation and improvement through continuous training
• Dynamic information retrieval for more efficient analysis

Technical implementation by our team

Our Software & Information Engineering department developed a machine learning algorithm that used image processing to classify parcels. The implementation took place in several steps:

1. Compilation of the training data

2. Selection of a suitable network architecture

3. Pre-processing the data and training the networks

4. Optimization of the training parameters for new training runs

5. Transfer of the trained network to industrial application

Classification was carried out using optical triggers. Challenges such as poor image quality or unidentifiable objects in the background initially led to misclassifications, which were gradually minimized through continuous training and optimization.

Results and benefits for logistics

The implementation of the machine learning approach led to:

• Improved automatic classification of packages
• Reduction of errors in the quantity forecast
• More efficient logistics processes through optimized information retrieval