Object Detection

Object detection with convolutional neural networks

Object Detection with CNNs

Why was this product built?

Optical surfaces are subject to strict regulations regarding particles, long imperfections (scratches), localised imperfections (wipers, holes, tool marks, coating defects) and edge chips. Many of these requirements are summarised in the ISO 10110-7 standard.
Often, in addition to the standardised classifications, there are also customer-specific requirements such as drawing marks, particles, coating defects and other defects that cannot be cleaned (saw marks, cracks). In this case, a simple and expandable integration of these specifications into surface inspection is necessary.
Available inspection systems have the disadvantage that the adaptation to new products is often associated with great effort. Furthermore, the reproducibility and repeatability of the classification is not always guaranteed.

What needs are satisfied with this product?

Neural networks can be adapted to the different requirements if the customer wishes so or if an already available network is not optimally suited. In this way a maximum of performance, expandability and accuracy is achieved.
Due to the fast and reliable classification we were able to halve the evaluation time and consequently double the output.

How has the problem been solved until today?

Many optical evaluation methods use classical image processing algorithms for classification and evaluation. In most cases, product adaptation requires a revision and extension of the algorithms.

object detection light one

Surface with regions of interest in which different criterias are enforced.

object detection light two

Surface that is hard to evaluate with conventional methods.


esp-engineering gmbh
Strahlholz 13
9056 Gais

E-Mail: info@esp-engineering.com
Phone: +41 71 791 0260

object detection zoomed dark

Zoomed in view. Only particles that match our pattern are reported.

object detection zoomed light

Zoomed in view. The surface area gets calculated and displayed.


Designed with ♥ by esp-engineering GmbH

Copyright © 2020 esp-engineering GmbH