Concerning the problem that the local feature and rotation invariant feature cannot be jointly paid attention to in traditional attention-based neural networks, a multi-branch neural network model based weakly supervised fine-grained image classification method was proposed. Firstly, the lightweight Class Activation Map (CAM) network was utilized to localize the local region with potential semantic information, and the residual network ResNet-50 with deformable convolution and Oriented Response Network (ORN) with rotation invariant coding were designed. Secondly, the pre-trained model was employed to initialize the feature networks respectively, and the original image and the above regions were input to fine-tune the model. Finally, the three intra-branch losses and between-branch losses were combined to optimize the entire network, and the classification and prediction were performed on the test set. The proposed method achieves the classification accuracies of 87.7% and 90.8% on CUB-200-2011 dataset and FGVC_Aircraft dataset respectively, which are increased by 1.2 percentage points, and 0.9 percentage points respectively compared with those of the Multi-Attention Convolutional Neural Network (MA-CNN) method. On Aircraft_2 dataset, the proposed method reaches 91.8% classification accuracy, which is 4.1 percentage points higher than that of ResNet-50. The experimental results show that the proposed method improves the accuracy of weakly supervised fine-grained image classification effectively.

At present, the face recognition methods based on deep learning have the problems of large model parameter size and slow feature extraction speed, and the existing face datasets have the problem of single pose, which cannot achieve good recognition effect in the actual face recognition task. Aiming at this problem, a multi-pose face dataset was established, and a lightweight multi-pose face recognition method was proposed. Firstly, the MTCNN (Multi-Task cascaded Convolutional Neural Network) algorithm was used by the method for face detection, and the high-level features included in the last network of MTCNN were used for face tracking. Then, the face pose was judged according to the positions of the detected face key points, the current face features were extracted by the neural network with ArcFace as loss function, and the current face features were compared with the face features of the corresponding pose in the face database to obtain the face recognition result. The experimental results show that the accuracy of the proposed method is 96.25% on the multi-pose face dataset, which is 2.67% higher than that on the face dataset with single pose. It shows that the proposed multi-pose face recognition method can effectively improve the recognition accuracy.

In order to solve the detection of ships with different types and sizes under complex marine environment, a real-time object detection algorithm based on deep learning was proposed. Firstly, a discriminant method between sharp and fuzzy such as rainy and foggy images was proposed. Then a multi-scale object detection algorithm based on deep learning framework of You Only Look Once (YOLO) v2 was proposed. Finally, concerning the character of remote sensing images of ships, an improved non-maximum supression and saliency partitioning algorithm was proposed to optimize the final detection results. The experimental results show that, on the dataset of ship detection in an open competition under complex sea conditions and meteorological conditions, the precision of the proposed method is increased by 16% compared with original YOLO v2 algorithm.

In view of feature selection and parameter optimization in Support Vector Machine (SVM) have great impact on the classification accuracy, an improved algorithm based on Particle Swarm Optimization (PSO) for SVM feature selection and parameter optimization (GPSO-SVM) was proposed to improve the classification accuracy and select the number of features as little as possible. In order to solve the problem that the traditional particle swarm algorithm was easy to fall into local optimum and premature maturation, the crossover and mutation operator were introduced from Genetic Algorithm (GA) that allows the particle to carry out cross and mutation operations after iteration and update to avoid the problem in PSO. The cross matching between particles was determined by the non-correlation index between particles and the mutation probability was determined by the fitness value, thereby new particles was generated into the group. By this way, the particles jump out of the previous search to the optimal position to improve the diversity of the population and to find a better value. Experiments were carried out on different data sets, compared with the feature selection and SVM parameters optimization algorithm based on PSO and GA, the accuracy of GPSO-SVM is improved by an average of 2% to 3%, and the number of selected features is reduced by 3% to 15%. The experimental result show that the features selection and parameter optimization of the proposed algorithm are better.

For the pseudo-linear output errorregressive systems whose identification model has the unknown variables in the information vector, this paper presented an auxiliary model based recursive least squares parameter estimation algorithm that was derived through constructing an auxiliary model and replacing the unknown inner variables with the outputs of the auxiliary model, but the effect was not good. Furthermore, through filtering the observation data with the estimated transfer function of the noise model and using the filtered data to estimate the parameters, the authors presented a data filtrating based recursive least squares parameter estimation algorithm. The simulation results show that the proposed algorithm can estimate the parameters of pseudo-linear output errorregressive systems effectively.