Radial Basis Function Neural Network sebagai Pengklasifikasi Citra Cacat Pengelasan

<p>Tingginya resiko kesalahan manusia dalam inspeksi visual untuk cacat pengelasan yang masih mengandalkan kemampuan manusia sulit untuk dihindari. Oleh sebab itu, penelitian ini mengusulkan sebuah klasifikasi cacat las visual dengan menggunakan algoritma <em>Radial Basis Function Neural Network</em> (RBFNN). Masukan RBFNN berupa citra las yang terdiri dari 5 (lima) kelas cacat las visual dan 1 (satu) kelas citra las normal. Citra las tersebut diproses terlebih dahulu menggunakan metode ekstraksi fitur <em>Fast Fourier Transform</em> (FFT) dan <em>Descreate Cosine Transform</em> (DCT). Hasil kedua metode ekstraksi fitur tersebut kemudian akan saling dibandingkan untuk mengetahui kinerja RBFNN. Hasil pengujian menunjukkan bahwa sistem dengan metode FFT-RBFNN dapat menggolongkan citra cacat las dengan akurasi sebesar 91.67% dan DCT-RBFNN sekitar 83.33% dengan jumlah neuron hidden layer sebanyak 15 dan parameter spread adalah 4.<em></em></p><p>Kata Kunci: <em>Radial Basis Function Neural Network</em> (RBFNN), FFT, DCT, cacat las, klasifikasi.</p><p align="center">Radial Basis Function Neural Network as a Weld Defect Classifiers<strong></strong></p><p><strong> </strong></p><p><strong>ABSTRACT</strong></p><p><em>The high risk of human error in visual inspection of welding defects that still rely on human capabilities is difficult to avoid. Therefore, this study proposes a classification of visual welding defects using the Radial Base Function Neural Network (RBFNN) algorithm. The RBFNN input is in the form of a welding image consisting of 5 (five) visual welding defect classes and 1 (one) normal welding image class. The weld image is processed first using the Fast Fourier Transform (FFT) and Descreate Cosine Transform (DCT) feature extraction methods. The results of these two feature extraction methods will be compared to find out the RBFNN performance. The test results show that the system with FFT-RBFNN method can classify the image of weld defects with an accuracy of 91.67% and DCT-RBFNN around 83.33% with the number of hidden layer neurons as much as 15 and the parameters of spread are 4.</em></p><p><em>Keywords: Radial Basis Function Neural Network (RBFNN), FFT, DCT, weld defect, classification.</em></p>

A Multi-exposure Image Fusion Method with Detail Preservation

In view of the problems of uneven exposure in the image acquisition and the serious loss of details in the traditional multi-exposure image fusion algorithm, a method of image fusion with details preservation is proposed. A weighted approach to multi-exposure image fusion is used, taking into account the features such as local contrast, exposure brightness, and color information to better preserve detail. For the purpose of eliminating the noise and interference, using the recursive filter to filter. Compared with other algorithms, the proposed algorithm can retain the rich detail information to meet the quality requirements of spot welding image fusion and has certain application value.

A modified welding image feature extraction algorithm for rotating arc narrow gap MAG welding

Purpose The purpose of this paper is to provide a modified welding image feature extraction algorithm for rotating arc narrow gap metal active-gas welding (MAG) welding, which is significant for improving the accuracy and reliability of the welding process. Design/methodology/approach An infrared charge-coupled device (CCD) camera was utilized to obtain the welding image by passive vision. The left/right arc position was used as a triggering signal to capture the image when the arc is approaching left/right sidewall. Comparing with the conventional method, the authors’ sidewall detection method reduces the interference from arc; the median filter removes the welding spatter; and the size of the arc area was verified to reduce the reflection from welding pool. In addition, the frame loss was also considered in the authors’ method. Findings The modified welding image feature extraction method improves the accuracy and reliability of sidewall edge and arc position detection. Practical implications The algorithm can be applied to welding seam tracking and penetration control in rotating or swing arc narrow gap welding. Originality/value The modified welding image feature extraction method is robust to typical interference and, thus, can improve the accuracy and reliability of the detection of sidewall edge and arc position.

Research on Edge Extraction System of the Laser Welding Image Based on ARM

The article researched on edge extraction system of the laser welding image based on ARM micro controller. The system can achieve accurate extraction on the edge of the laser welding pool, and reduce the system cost effectively while satisfying the real-time constraints. With STM32f103 controller for the system as the core, the outside are image storage and display module. The image processing methods of gray-scale transformation, image smoothing, threshold segmentation and edge extraction are realized in ARM, and the processed images are displayed in LCD. The experimental results show that the system researched in this paper can extract color molten pool image edge accurately without distortion and better reflect the relationship between welding parameters and weld pool geometry size, and provide the basis for building real-time control system of the laser welding quality.