Image Denoising using Deep Auto-Encoder Network for Production Monitoring in Real-Time
Sujay Jadhav, Pooja Kulkarni · 2021 International Conference on Advances in Electrical, Computing, Communication and Sustainable Technologies (ICAECT) · 2021
Industrial operations today are being smoothed by monitoring the entire process using latest Image Processing (IP) techniques, especially the bottlenecks in the process. The challenges that need to be addressed in order to commercialize and optimize these solutions are equally growing, particularly in image monitoring of the machining and assembly lines in production plants as well as mechanical workshops. One such hindrance in monitoring the jobs on a Lathe machine or a Computer Numerical Control (CNC) machine is the motion blur noise introduced in the images due to the real-time on field vibrations within the setup. In this paper, this problem is addressed with Image denoising in real time frames. Many practical solutions have been addressing the problem of Image denoising or Image reconstruction using different Computer Vision (CV) techniques in the last two decades. Herein, we propose Image restoration of such blurred image using a specific architecture of Deep Autoencoder network with Convolutional Neural layers so as to obtain a fully constructed image without any prior knowledge of the corresponding clean image. For this we have used an averagely deep encoder-decoder neural network to minimize the typical motion blurring noise that gets introduced in the input image captured by the camera setup on the production lines. This particular technique eliminates the need to identify the type of noise, its intensity or any other statistical characteristics. The results evidently show that the reconstructed images are visually convincing, with large improvements retaining the blur image details.