Physics Maths Engineering
Attention Inspiring Receptive-Fields Multi-Task Network via Self- supervised Learning for Violence Recognition
Suyuan Li,
Xin Song
Abstract
Abstract Generally, a large amount of training data is essential to train deep learning model for obtaining more accurate detection performance in computer vision domain. However, to collect and annotate datasets will lead to extensive cost. In this letter, we propose a self-supervised auxiliary task to learn general videos features without adding any human-annotated labels, aiming at improving the performance of violence recognition. Firstly, we propose a violence recognition method based on convolutional neural network with self-supervised auxiliary task, which can learn visual feature for improving down-stream task (recognizing violence). Secondly, we establish a balance-weighting scheme to solve the crucial problem of balancing the self-supervised auxiliary task and violence recognition task. Thirdly, we develop an attention receptive-field module, indicating that the proper use of the spatial attention mechanism can effectively expand the receptive fields of the module, further improving semantically meaningful representation of the network. To evaluate the proposed method, two benchmark datasets have been used, and better performance is shown by the experimental results comparing with other state-of-the-art methods.
Key Questions
What is the focus of the study?
The study focuses on enhancing image-based malware detection by applying α-cuts to binary visualizations of malicious binaries, aiming to improve color and pattern segmentation and achieve sparse image representations.
How are α-cuts utilized in this research?
In this research, the R, G, and B color values of each pixel are considered as respective fuzzy sets. α-cuts are then applied as a defuzzification method across all pixels, converting them into sparse matrices of 0s and 1s, thereby enhancing color and pattern segmentation.
What methodology was used to evaluate the proposed approach?
The proposed method was tested on various dataset sizes and evaluated using hyperparameterized ResNet50 models. The performance metrics included accuracy, precision, recall, and f-score to assess the effectiveness of the approach.
What were the key findings of the study?
The study found that for larger datasets, the sparse representations of intelligently colored binary images achieved through α-cuts can surpass the performance of unprocessed images. Specifically, the method achieved 93.60% accuracy, 94.48% precision, 92.60% recall, and a 93.53% f-score.
What is the significance of this research in the field of image processing?
This research is significant as it is the first to apply α-cuts in image processing for malware detection. The findings suggest that α-cuts provide an important contribution to handling challenging datasets and can be integrated into image-based Intrusion Detection Systems (IDS) and other demanding real-time applications.
