A multi-view object detection approach based on deep learning is proposed in this paper. Classical object detection methods based on regression models are introduced, and the reasons for their weak ability to detect small objects are analyzed. To improve the performance of these methods, a multi-view object detection approach is proposed, and the model structure and working principles of this approach are explained. Additionally, the object retrieval ability and object detection accuracy of both the multi-view methods and the corresponding classical methods are evaluated and compared based on a test on a small object dataset. The experimental results show that in terms of object retrieval capability, Multi-view YOLO (You Only Look Once: Unified, Real-Time Object Detection), Multi-view YOLOv2 (based on an updated version of YOLO), and Multi-view SSD (Single Shot Multibox Detector) achieve AF (average F-measure) scores that are higher than those of their classical counterparts by 0.177, 0.06, and 0.169, respectively. Moreover, in terms of the detection accuracy, when difficult objects are not included, the mAP (mean average precision) scores of the multi-view methods are higher than those of the classical methods by 14.3%, 7.4%, and 13.1%, respectively. Thus, the validity of the approach proposed in this paper has been verified. In addition, compared with state-of-the-art methods based on region proposals, multi-view detection methods are faster while achieving mAPs that are approximately the same in small object detection.

The bank telemarketing dataset was collected by a Portuguese retail bank when selling a bank long-term deposits product. In order to predict the success of bank telemarketing, an more accurate approach was proposed. Firstly, the dataset was analyzed and processed. The features value of the dataset could be divided into two kinds: continuous variable and categorical variable. The two kinds of variable had been processed with different methods. The continuous variables were transformed to discrete features and then were normalized. The categorical variables were coded with two different methods according to whether they were belong to the ordered categories. The missing value of features had to been processed with two different methods depending on the examples number of the missing value. Secondly, the three data mining models which were support vector machine, neural network and decision trees were adopted to predict the result of the processed bank telemarketing dataset. The classification and regression trees algorithm was adopted in the decision trees model. The BFGS algorithm had been used when the neural network model was programming. The simple sequential minimal optimization (SMO) algorithm was adopted in the support vector machine model. All the three models were executed using python 2.7 and conducted in a Windows7 PC, with an Intel Cerelon G1630 2.80 GHz processor. The results of three models were analyzed and compared. The decision trees presented the best results and cost the least time, and its accuracy, precision, recall value and AUC area of the were all equal to 1.0. It showed that decision trees could predict 100% correct rate. The accuracy, recall value and AUC area of the neural network were better than those obtained by the support vector machine (improvement of 1.5 pp, 2.7 pp and 5 pp respectively), but the neural network cost almost 4.4 times more time than the support vector machine. It showed that not only choosing the data mining models but also processing datasets in advance played a key role in obtaining better prediction results when datasets needed classifying.

Because the / k-Nearest Neighbors(k-NN) algorithm does not take into account different weights values among each feature of data sets, it sometimes achieves low accuracy. In order to increase the k-NN algorithm accuracy, the genetic algorithm-based (GA-based) weight-tunning method is proposed. When the features have been given the same weights values, the k-NN algorithm with EU distance metric is introduced firstly. Secondly, the weighted k-NN algorithm is obtained, when considering different weights values among features. Thirdly, the steps of GA-based weight-tunning method are given in detail. Finally, the experimental tests are running with the real data sets about the students' knowledge status about the subject of Electrical DC Machines. The experimental results show that the weighted k-NN algorithm achieves classification accuracy of 95.2%, whereas the classification accuracy of k-NN algorithm is 86.2%. The GA-based weight-tunning approach for features of data set could play an essential role in improving the k-NN classifier accuracy.