Search

Multi-objective evolutionary algorithm based on decomposition (MOEA/D) is effective to solve most multi-objective optimization problems (MOPs) in the past 20 years. However, the algorithm MOEA/D with constant weight vectors has bad performance in solving several MOPs with discontinuous Pareto front (PF). This paper analyses the limitations of the constant weight vectors in MOEA/D and explains the necessity of adjusting the weight vectors in the processing. This paper proposes a weight vector adjustment method for MOEA/D (MOEA/D-WVA). It deletes the weight vectors which have bad search direction, and adds some new weight vectors in the processing. Experimental studies are conducted on several MOPs with discontinuous PF to compare the MOEA/D-WVA with other state-of-the-art multi-objective optimization algorithms in solving those MOPs with complex PF. The results show MOEA/D-WVA performs better than other algorithms on those MOPs with discontinuous PF.

Multi-objective evolutionary algorithm based on decomposition (MOEA/D) has achieved great success in the field of evolutionary multi-objective optimization. It decomposes a multi-objective optimization problem into a number of scalar optimization sub-problems. Each sub-problem is optimized by using information from its neighboring sub-problems. Therefore, the neighborhood size of each sub-problem plays an important role in MOEA/D. Different neighborhood sizes are tested in this paper. Experimental results demonstrate that larger neighborhood size helps achieve better convergence and diversity with more CPU time and vice versa. MOEA/D uses constant neighborhood size during the whole process, and it is difficult to balance the convergence, diversity and running time. Therefore, this paper propose an algorithm based on MOEA/D. The algorithm adjusts the neighborhood size dynamically in different generations and different sub-problems to reduce the running time while the convergence and diversity of this algorithm are similar or better than other state-of-the-art algorithms. Compared to the original MOEA/D, experimental results show that adjusting the neighborhood size dynamically is a good way to reduce the running time significantly while maintaining the convergence and diversity. Furthermore, the algorithm proposed in this paper is compared with five state-of-the-art algorithms based on MOEA/D. The experimental results show that the proposed algorithm outperforms the others in efficiency while performs similarly in convergence and diversity.

Short-wave radio is an indispensable long-distance means of communication, among which Morse signals, which rely on simplicity and efficiency, plays an import role in military and civilian applications. Automatic Morse detection and recognition have been researched for several years, but some thorny problems in actual communication always restrict the performance of methods. In this article, by introducing deep learning technology, we propose a network named MorseNet that can simultaneously locate and decode Morse signals in the spectrogram. MorseNet uses shared convolutions to extract shared features for both the detection and recognition branches. The detection branch regresses bounding boxes based on signal centerlines, and the recognition branch decodes Morse fragments cropped from feature maps by a convolutional recurrent neural network (CRNN). The losses of two branches are combined to implement the end-to-end training. Experimental results on four “simulated Morse + real background” datasets demonstrate that the proposed method achieves state-of-the-art performance in both detection and recognition, and it effectively improves four problems that have long been troublesome in accomplishing the tasks. Furthermore, the joint training strategy and architecture give MorseNet advantages over its two-stage deployment in terms of accuracy, speed, and model size.