نوع مقاله : مقاله پژوهشی (کاربردی)
نویسندگان
گروه مهندسی کامپیوتر، دانشگاه فنی و حرفهای، تهران، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Deep learning is a subset of machine learning that is widely used in the field of artificial intelligence, such as natural language processing and machine vision. As a subset of machine vision, image segmentation is one of the most common steps in digital image processing, which divides a digital image into different segments. In this research, a new method based on deep learning for image segmentation is presented. The "Adaption Ahead" algorithm was introduced and used as a new optimization algorithm to optimize the proposed model. In previous optimization algorithms, the most important factor in reducing accuracy is extracting low-level features of images and not reducing the semantic distance between human perception and these features. In this study, hierarchical and deep feature extraction from images was done with the help of deep learning. The "Adaption Ahead" optimization algorithm, in which a deep model based on a convolutional neural network is used, has extracted higher-level features and achieved optimal accuracy. By using the Nestrov technique in calculating the gradient by the proposed algorithm, the best result, i.e. 91.1 accuracy, was obtained for the Dice similarity measure. Another advantage of this algorithm over other methods is using uncomplicated calculations. The comparison of the proposed optimization algorithm with other commonly used methods shows the improvement in the performance of this network on relatively large data sets. Also, the more accurate performance of this network, as a result of its hierarchical and deep extraction, is compared to other methods.
کلیدواژهها [English]