Iris recognition is one of the most representative identification technologies in biometric recognition, which is widely used in various fields. Recently, many deep learning methods have been used in biometric recognition, owing to their advantages such as automatic learning, high accuracy, and strong generalization ability. The deep convolutional neural network (CNN) is the mainstream method of image processing widely used in many domains, but it has poor anti-noise capacity in image classification and is easily affected by slight disturbances. CNN also needs a large number of samples for training. The recent capsule network not only has high recognition accuracy in classification tasks but can also learn part-whole relationships, increasing the robustness of the model. Furthermore, it can be trained using a small number of samples. In this paper, we propose a deep learning method based on the capsule network architecture in iris recognition. The structure detail of the network is adjusted, and we provide a modified routing algorithm based on the dynamic routing between two capsule layers to make this technique adapt to iris recognition. Migration learning makes the deep learning method available even when the number of samples is limited. Therefore, three state-of-the-art pretrained models, VGG16, InceptionV3, and ResNet50, are introduced. We divide the three networks into a series of subnetwork structures according to the number of their major constituent blocks. They are used as the convolutional part to extract primary features, instead of a single convolutional layer in the capsule network. Our experiments are conducted on three iris datasets, JluIrisV3.1, JluIrisV4, and CASIA-V4 Lamp, to analyze the performance of different network structures. We also test the proposed networks in simulated strong and weak light environments, showing that the networks with capsule architecture are more stable than those without.

Indoor positioning technologies are of great use in GPS-denied areas. They can be partitioned into two types of systems—infrastructure-free based and infrastructure-dependent based. WiFi based indoor positioning system is somewhere between the infrastructure-free and infrastructure-dependent systems. The reason is that in WiFi based systems, Access Points (APs) as pre-installed infrastructures are necessary. However, the APs do not need to be specially installed, because WiFi APs are already widely deployed in many indoor areas, for example, offices, malls and airports. This feature makes WiFi based indoor positioning suitable for many practical applications. In this paper, a seq2seq model based, deep learning method is proposed for WiFi based fingerprinting. The model can learn from different length of training sequences, and thus can exploit the context information for positioning. The context information denotes the information contained in the sequence, which can help finding the correspondences between RSS fingerprints and the coordinate positions. A simple example piece of context information is human walking routine (such as no sharp turns). The proposed method shows an improvement with an open source dataset, when compared against deep learning based counterpart methods.

Radio frequency communication technology has not only greatly improved public network service, but also developed a new technological route for indoor navigation service. However, there is a gap between the precision and accuracy of indoor navigation services provided by indoor navigation service and the expectation of the public. This study proposed a method for constructing a hybrid dual frequency received signal strength indicator (HDRF-RSSI) fingerprint library, which is different from the traditional RSSI fingerprint library constructing method in indoor space using 2.4G radio frequency (RF) under the same Wi-Fi infrastructure condition. The proposed method combined 2.4G RF and 5G RF on the same access point (AP) device to construct a HDRF-RSSI fingerprint library, thereby doubling the fingerprint dimension of each reference point (RP). Experimental results show that the feature discriminability of HDRF-RSSI fingerprinting is 18.1% higher than 2.4G RF RSSI fingerprinting. Moreover, the hybrid radio frequency fingerprinting model, training loss function, and location evaluation algorithm based on the machine learning method were designed, so as to avoid limitation that transmission point (TP) and AP must be visible in the positioning method. In order to verify the effect of the proposed HDRF-RSSI fingerprint library construction method and the location evaluation algorithm, dual RF RSSI fingerprint data was collected to construct a fingerprint library in the experimental scene, which was trained using the proposed method. Several comparative experiments were designed to compare the positioning performance indicators such as precision and accuracy. Experimental results demonstrate that compared with the existing machine learning method based on Wi-Fi 2.4G RF RSSI fingerprint, the machine learning method combining Wi-Fi 5G RF RSSI vector and the original 2.4G RF RSSI vector can effectively improve the precision and accuracy of indoor positioning of the smart phone.

Unlike score level fusion, feature level fusion demands all the features extracted from unimodal traits with high distinguishability, as well as homogeneity and compatibility, which is difficult to achieve. Therefore, most multimodal biometric research focuses on score level fusion, whereas few investigate feature level fusion. We propose a face–iris recognition method based on feature level fusion. We build a special two-dimensional-Gabor filter bank to extract local texture features from face and iris images, and then transform them by histogram statistics into an energy-orientation variance histogram feature with lower dimensions and higher distinguishability. Finally, through a fusion-recognition strategy based on principal components analysis and support vector machine (FRSPS), feature level fusion and one-to-n identification are accomplished. The experimental results demonstrate that this method can not only effectively extract face and iris features but also provide higher recognition accuracy. Compared with some state-of-the-art fusion methods, the proposed method has a significant performance advantage.

In recent years, fire accidents in petrochemical plant areas and dangerous goods storage ports in China have shown a trend of frequent occurrence. Toxic and harmful gases are diffused in the scenes of these accidents, which causes great difficulties for fire fighting and rescue operations of fire fighting forces, and consequently, casualties of firefighters often occur. In order to ensure the safety of firefighters in such places, this paper designs a monitoring system of toxic and harmful gases specially used in fire fighting and rescue sites of fire forces, and establishes the transmission network, monitoring terminal and data processing software of the monitoring system of toxic and harmful gases, establishing the danger model of the monitoring area of toxic and harmful gas-monitoring terminal, and the danger model of fire fighters’ working area, fusing the field toxic and harmful gas data, terminal positioning data, and field environmental data, designing the data structure of the input data set and the network structure of the RNN cyclic neural network model, and realizing the dynamic early warning of toxic and harmful gases on site.