At present, the research hotspot of small target detection mainly focuses on the methods based on deep learning, and such algorithms still have some problems that need to be solved, such as the foreground-background class imbalance and the poor performance of multi-scale detection. For the foreground-background class imbalance, the Squared Cross Entropy (SCE) loss function is proposed here to help solve the problem. Meanwhile, as Feature Pyramid Networks (FPN) is a powerful means to deal with multi-scale detection problems, a new Dense FPN structure is designed based on FPN. The Dense FPN removes the up-sampling process in FPN, and after each feature extraction layer, a continuous convolutional layer with a decreasing number of layers is added. According to the experimental results, Dense FPN outperform the original FPN on various evaluation indicators like AP S , AP M and AP L , showing the excellent performance of the Dense FPN in dealing with multi-scale detection problems.

It is known that the optimization of the Earth-Moon low-energy transfer trajectory is extremely sensitive with the initial condition chosen to search. In order to find the proper initial parameter values of Earth-Moon low-energy transfer trajectory faster and obtain more accurate solutions with high stability, in this paper, an efficient hybridized differential evolution (DE) algorithm with a mix reinitialization strategy (DEMR) is presented. The mix reinitialization strategy is implemented based on a set of archived superior solutions to ensure both the search efficiency and the reliability for the optimization problem. And by using DE as the global optimizer, DEMR can optimize the Earth-Moon low-energy transfer trajectory without knowing an exact initial condition. To further validate the performance of DEMR, experiments on benchmark functions have also been done. Compared with peer algorithms on both the Earth-Moon low-energy transfer problem and benchmark functions, DEMR can obtain relatively better results in terms of the quality of the final solutions, robustness, and convergence speed.