Search

Skin cancer is a perilous kind of cancer caused by damaged DNA and it leads to death. This damaged DNA causes uncontrolled proliferation of cells. Even though, the image analysis of lesions is highly difficult due to light reflections from skin surface, fluctuations at color lighting, variety of lesions' forms and sizes in skin cancer. Because of these issues, automatic recognition of skin cancer accurateness is decreased. Therefore, a Graph Convolutional Neural Network (GCNN) by ResNet 152 Transfer Learning Architype optimized with Chameleon Swarm Optimization Algorithm (GCNN-ResNet 152 TL-CSOA) is proposed at this manuscript for enhancing skin cancer detection with classification in medical image processing. Initially, the input images are taken from International Skin Imaging Collaboration (ISIC) of dermoscopic skin cancer imagery data set. Afterward, the input images are pre-processed utilizing trilateral filter method for removing noise. The pre-processed output is supplied to the process of feature extraction. Here, image features, like morphologic, gray scale statistic and Haralick texture features are extracted by Gray-Level Co-Occurrence Matrix window adaptive approach (GLCM-WAA) technique. After that, the GCNN-ResNet 152 TL classifies the skin cancer images into Actinic Keratosis, Basal Cell Carcinoma, Malignant Melanoma and Squamous Cell Carcinoma. Additionally, GCNN-ResNet 152 TL weight parameters is tuned by Chameleon Swarm Optimization Algorithm (CSOA). The simulation process is executed at Python tool. From simulation, the proposed approach attains 23.34%, 12.03%, 21.42% improved accuracy and 18.23%, 21.23%, 14.56% higher sensitivity compared with existing approaches. [ABSTRACT FROM AUTHOR]

In wireless communications, cognitive radio systems are widely used. But in real time applications several interferences occur due to multiple input multiple output systems. The main reason for interference is imperfect spectrum sensing for channel allocation to secondary users. Secondary users try to access primary channels and switch to another frequency when primary user needs access. During this phenomenon interference come in to the picture between the two users. In such a scenario, beamforming provides information security and avoids interference. In this work, a new variable step size based adaptive learning is considered for beamforming. Variable step size scheme is used for distributed estimation in adaptive networks using diffusion concept. This learning scheme iteratively update the step size and reduces the power of error. The performance of proposed diffusion variable step size LMS (DVSLMS) algorithm is studied in multipath environment. The convergence analysis is carried to prove the ability of adaptive learning methods. The experimental results are presented in terms of convergence rate and number of active taps at threshold points 0.1, 0.5 and 1.0. [ABSTRACT FROM AUTHOR]