Human-machine conversation plays an important role in natural language processing and artificial intelligence. Human-machine conversation can be divided into the question answering system, task-oriented conversation, and chatting system according to the purpose of use. Among them, the chatting system usually requires higher personification. Based on the sequence transformation model of the long short-term memory network, the topic network is introduced in this study to explicitly extract the scene and topic information from the conversation, and this higher-level feature, which does not change with the word order, is inputted to the structure of the conversation model to guide the decoding and prediction processes together with the attention mechanism. Because of the difficulty of obtaining the topic information in advance, the topic network is modeled as an unsupervised learning structure. Thus, the training process needs to be divided into three steps. The experimental results show that the model can significantly improve the quality of the chatting system with appropriate training methods and structural parameters.

A position tracking control method was designed with an attitude extraction algorithm for a tailsitter vertical take-off and landing (VTOL) unmanned aerial vehicle (UAV) subjected to time-varying crosswind disturbances. A mathematical model was constructed to predict the relationship between the crosswind disturbance and the aircraft's attitude. An attitude extraction algorithm was then developed using the aircraft's roll angle which significantly reduces the effect of the disturbance. A backstepping control method was used with a saturated function technique to account for the thrust boundedness with a PD plus feedforward control method for the attitude tracking control. The unknown derivatives of the desired attitude caused by the time-varying crosswind disturbance were resolved by a command filter. Tests show that the tracking error can be made arbitrarily small as long as the command filter frequency is sufficiently large.

The centroid location of a near infrared star always deviates from the real center due to the effects of surrounding radiation. To determine a more accurate center of a near infrared star, this paper proposes a method to detect the star’s saliency area and calculate the star’s centroid via the pixels only in this area, which can greatly decrease the effect of the radiation. During saliency area detection, we calculated the boundary connectivity and gray similarity of every pixel to estimate how likely it was to be a background pixel. Aiming to simplify and speed up the calculation process, we divided the near infrared starry sky image into super pixel maps at multi-scale by Simple Linear Iterative Clustering (SLIC). Second, we detected the saliency map for every super pixel map of the image. Finally, we fused the saliency maps according to a weighted coefficient that is determined by the least square method. For the images used in our experiment, we set the multi-scale super pixel numbers to 100, 150, and 200. The results show that our method can obtain an offset variance of less than 0.27 for the center coordinates compared to the labelled centers.