The Proposal of Customized Convolutional Neural Network Using for Image Blur Recognition
This paper describes one of the parts of our research topic devoted to the image recognition and computer vision. In the field of automation, it is not new, that industry lines are under controls of camera systems. It is necessary to have a clear and visi
- PDF / 1,202,889 Bytes
- 8 Pages / 439.37 x 666.142 pts Page_size
- 94 Downloads / 198 Views
Abstract. This paper describes one of the parts of our research topic devoted to the image recognition and computer vision. In the field of automation, it is not new, that industry lines are under controls of camera systems. It is necessary to have a clear and visible image for automatic detection of various damages, fails, alarms, objects and right in time control. We divided our research into several stages. In the first stage, we are dealing with the recognition of the blur in the image. We are dealing with classification problem whether the image is blured or in focus. In the initial steps of our research, we have created a dataset of images, which is divided into four categories. In this paper we describe suitable methods to solve this task, we discuss the pros and cons and we also propose our own classification model based on artificial neural networks. Keywords: Image processing
Convolutional neural networks ReLU
1 Introduction Machine learning methods are used more and more in the field of automation. These methods are used to make the process control more effective and also safe. Artificial intelligence methods and algorithms can solve various complex tasks, can be also used in the decision process and they also can ensure safety of the controlled processes. Each process is becoming automized and more effective. Even the detection of various faults, fails and damages is possible with the use of machine learning methods combined with the camera system. This type of automatic fault detection, when the production line is under the control is very convenient. The issue of fitting such systems into production is becoming very current topic. In this issue of deployment of a camera system into the production, it is important to solve several problems. Authors in [1] are dealing with the camera angles. In paper [2] authors are devoting to getting the high-quality image for scanning surface. In our research, we chose the problem of the quality of the image. It is very important to have a clear and visible image for further use for fault detection. We are
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2020 R. Silhavy et al. (Eds.): CoMeSySo 2020, AISC 1294, pp. 821–828, 2020. https://doi.org/10.1007/978-3-030-63322-6_69
822
D. Borkin et al.
solving the classification problem to classify images into blurred or in focus. For this purpose, we use neural networks.
2 Background and Methods In our research, we have a dataset of images. The dataset itself was not big, because we have decided to create the dataset ourselves. We have created 4 images for each object. These 4 images represent 4 classes/categories. First class represents sharp images. The second class represents images slightly out of focus and for each of the next classes the image is more out of focus. Images in the last 4-th class were totally out of focus, which means they were significantly blurred. The images in the training set were in the RGB format and the resolution for each image was 48
Data Loading...
