Recurrent nonsymmetric deep auto encoder approach for network intrusion detection system

YOLOv3, the third edition of the YOLO family, performs well on object detection, but using it for real-time vehicle and object detection on unmanned vehicles with limited computing capacity remains a very challenging task YOLOv3 has a high computational complexity. The main objective is to develop network architecture for vehicle and object detection based on YOLOv3. The total work will be broken down into three phases.
Firstly, to reduce the model size and computing complexity, we introduce L1 regularization to the batch normalization layer, which allows us to recognize and remove distracting channels and layers. Secondly, to reduce the missed detection in
crowded scenes and locate targets better, the Merge Soft-NMS which merges the bounding boxes with high overlap is designed based on Soft-NMS. Thirdly, considering the obvious aspect ratio of vehicle and objects, the anchor boxes which are designed
based on multi-class is redesigned for better vehicle and object matching and localization in YOLOv3. In the experiment, compared with SINGLE SHOT DETECTION (SSD) and YOLOv3 which performs well on detection accuracy and speed is effective and compact for vehicle and object detection.