The prevalence of skin cancer, specifically melanoma, constitutes a significant global health concern, thus giving rise to intricate detection challenges that demand immediate attention and comprehensive solutions. In this study, we investigate the application of deep learning models for melanoma detection. Five pre-trained models, including VGG-16, ResNet50, InceptionV3, DenseNet-121, and Xception, are evaluated through a series of experiments. The models undergo the same training process with transfer learning, freezing all layers and modifying the classification layer. The experiments reveal that ResNet50 consistently outperforms the other models, demonstrating superior accuracy, precision, recall, and F1 score. Notably, ResNet50 exhibits exceptional accuracy and F1 score, achieving around 93% in both. This study sheds light on the potential use of deep learning in enhancing melanoma diagnosis and underscores the need for robust and accurate classification systems for early detection and effective treatment of skin cancer.
International transport security policy requires the baggage screening. In airports and railway stations, these operations are processed manually with the help of X-ray or computed tomography machines. There is a need for an automatic system which could reduce the time of the screening process and possibly, increase the accuracy of the detections. More than that, there is a demand for developing and evaluating methodologies for learning on 3D image-like data, which has been addressed only recently, mostly in the field of medical imaging. The main objective of this research is to develop a framework for object detection in 3D computed tomography scans for high-throughput security applications. In this paper, a literature review on the topic of 3D image recognition is presented, and a transfer learning approach is evaluated on the security risk detection task in X-ray images.
The work done in this paper provides solutions to the collision avoidance of multiple point-mass robots in the presence of multiple spherical and cylindrical obstacles in three-dimensional space. A set of autonomous controllers is derived using the architecture of the Lyapunov-based Control Scheme (LbCS) to achieve this task. We implement the leader-follower strategy via the split-rejoin method here, where we choose a leader and the rest of the flock or agents behave with respect to the leader to achieve its predetermined task. In addition, in the presence of fixed obstacles, the whole formation will split to avoid obstacles present in its path using the Minimum Distance Technique and rejoin at a safer place, regaining its pre-defined formation with respect to the leader. The Minimum Distance Technique allows a point-mass agent to avoid the walls of obstacles by approaching them from the closest possible distance in each unit of time. The formation will also avoid inter-collision, which is defined as moving obstacles in this case. Using the Lyapunov-based control scheme, a set of non-linear acceleration-based controllers will be derived such that the desired task is easily completed. The controllers will ensure that the formation travels in a well-planned manner and successfully converges on its targets while avoiding obstacles. To prove the efficiency and simplicity of the nonlinear acceleration-based controllers derived, computer simulations of virtual scenarios have been performed and given in the simulation results and discussion section. The paper concludes with a conclusion and recommendations for future work in this field.
Utilizing classical convolutional networks results in lackluster performance in certain classification tasks. To address this problem, recent solutions add extra layers or sub-networks to increase the classification performance of existing networks. More recent methods employ multiple networks coupled with varying learning strategies. However, these approaches demand larger memory and computational requirement due to additional layers, prohibiting usage in devices with limited computing power. In this paper, we propose an efficient convolutional block which minimizes the computational requirements of a network while maintaining information flow through concatenation and element-wise addition. We design a classification architecture, called Half-Append Half-Add Network (HAHANet), built using our efficient convolutional block. Our approach achieves state-of-the-art accuracy on several challenging fine-grained classification tasks. More importantly, HAHANet outperforms top networks while reducing parameter count by at most 54 times. Our code and trained models are publicly available at https://github.com/dlsucivi/HAHANet-PyTorch .
Machine learning applications in remote sensing often require a labour-intensive feature engineering step, if only a small number of samples is available and transfer learning is not applicable. Here, we are introducing the concept of Spatial Variation Sequences, which allows to apply methodologies from automated time-series feature engineering to remote sensing applications of static images. The presented example application detects swimming pools from four-channel satellite images with an $$F_{1}$$ -score of 0.95, by generating spatial variation sequences from a modified swimming pool index. The automated feature engineering approach reduced the dimensionality of the classification problem by 99.7%. A more traditional approach using transfer learning on pre-trained Convolutional Neural Networks (CNN) was evaluated in parallel for comparison. The CNN approach boasted a higher performance of $$F_{1}$$ -score of 0.98 but required the use of pre-trained weights. The comparable performance of the FE and CNN approach demonstrates that time-series feature extraction is a valuable alternative to traditional remote sensing methods in the presence of data scarcity or the need of significant dimensionality reduction.
