UAV path planning is an important role in the UAV autonomy, and has become a hotspot. In actual combat scenarios, multiple UAVs are needed to strike multiple targets in succession and eventually return to base. But, multiple UAV path planning for the multiple targets is a challenge for the optimal path quickly. To solve the problem, a double precision method is designed in this paper. ant colony algorithm and an artificial bee colony algorithm are used to enhance the exploration capability in the low precision search phase and the exploitation capability in the high precision search phase, respectively. Simulation results show that the method can avoid fall into a local optimum, and effectively improve the efficiency of the path planning for the multi-UAV multi-mission objective problem.

In this paper, we propose AudioWrite, an acoustic-based handwriting recognition system. It utilizes the built-in microphone to capture sounds produced by the finger sliding on the surface of the object to recognize handwritten characters. We first design a segmentation algorithm to detect handwritten sounds in real time and accurately segment character signals. Then we extract time-frequency features from sound signals by performing short-time Fourier transform (STFT). Moreover, in order to obtain sufficient training samples at low cost to build an effective classification model, we adopt Generative Adversarial Network (GAN) to generate synthetic data with only a small set of real-world data. Since mobile devices are usually limited in storage and computing resources, we use the lightweight ShuffleNetV2 model to recognize handwritten characters. We implement AudioWrite as an Android application and conduct experimental evaluations. The results demonstrate that AudioWrite can recognize handwritten characters (10 numbers and 26 uppercase letters) in real time with an overall accuracy of 92.2%.

Obstacle detection is very important for people with visual impairment to avoid collision. However, traditional methods usually suffer from extra devices or susceptibility to light. To address these issues, we propose an obstacle detection system, ObstaDetect, which senses echoes in the environment to identify different obstacles in a independent manner on smartphones leveraging acoustic signals. It detects whether there is an obstacle ahead in real time and detects whether we are approaching an obstacle in a dynamic manner. First, the smartphone emits ultrasound, and the built-in microphone records the echo reflected by the obstacle. After that, we use band pass filter to remove noise. Finally, we build a Convolutional Neural Network (CNN) to extract the features of obstacles to distinguish them. In addition, Generate Adversarial Network (GAN) is used to generate large amounts of data similar to the original samples. Ultimately, ObstaDetect is implemented on a smartphone without additional hardware and devices. The experimental results show that ObstaDetect achieves a high accuracy of 96%.

Sign language is a bridge for communication be-tween people with normal hearing and impaired hearing. How-ever, only a few people with normal hearing have an understanding of sign language. Thus, Sign Language Recognition (SLR) has attracted significant interest from both industry and academia. Prior traditional methods have certain limitations such as light condition and extra hardware. To address these issues, this study aims to recognize sign languages using acoustic signals on a smartphone. We design an SLR system called SignGest, which captures user's sign language gestures with built-in microphones. Afterwards, we build a Convolutional Neural Network(CNN) model to extract features of different gestures to distinguish them. Furthermore, we use a Deep Convolutional Generative Adversarial Network(DCGAN) to generate abundant training data that look no different from the real samples. We handle various challenges including effective gesture segmentation and training data collection. Finally, SignGest is implemented on a server and an Android smartphone with built-in microphones and speakers without any extra hardware or infrastructure equipment. When users perform the sign language gestures before the smartphone, SignGest can recognize each of them and print the result. Through extensive experiments, our results show that SignGest can achieve robust and satisfactory performance.

The path planning of Unmanned Aerial Vehicle (UAV) is a real-world optimization problem, and even develops into a hard optimization problem with many objectives and constraints when UAVs work in a complex environment. In a complex environment, the resulting constraints can lead to decrease the quantity of the feasible solutions, so that can bring difficulties to plan routes for UAVs. Therefore, it is necessary to design a high-quality planner for a UAV in a complex environment. In this work, we have proposed a Multi-Strategy Fusion Differential Evolution algorithm (MSFDE). The proposed algorithm integrates the multi-population strategy, the novel self-adaptive strategy and the ensemble of the interactive mutation strategy in order to balance the exploitation and exploration capabilities. The multi-population strategy is used to divide the whole population into the three indicator subpopulations and a reward subpopulation for maintaining the diversity of the whole population; the novel self-adaptive strategy is introduced to control the parameters F and CR based on the teaching-learning-based optimization method; the ensemble of the interactive mutation strategy is to exchange the information among the three indicator subpopulations on each generation for boosting the population diversity. The constraints in the UAV path planning are transformed into the objective functions by the linear weighted sum method. Scenario 1, 2, 3, and 4 are designed with different complex level, and other eight algorithms are introduced to be compared with MSFDE. The simulation results confirm that MSFDE has an outstanding performance for the UAV three-dimensional path planning in the complex environment.