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.

Aiming at the contradiction between accuracy and speed in industrial object detection, this paper proposes an industrial object detection method based on improved CenterNet. The improved method uses ResNet-50 as the Backbone to boost detection speed, and an upsampling layer is added to the feature processing network to improve detection accuracy. The expermient results show that the mAP of the improved method reaches 87.41 %, which is 3.44% higher than the CenterNet-Res101 method, and the detection speed reaches 31 FPS, which is 4 FPS faster than the CenterNet-Res101 method.

In the Cost-Per-Mille (CPM) advertising business scenario that adopts the guaranteed delivery model, ad publisher predicts the future available impressions inventory of each advertising placement under the constraints of each target audience based on the historical impressions data, so as to sell advertising resources reasonably. However, historical data threatens user and publisher privacy during the Sharing-Aggregation-Analysis procedure. In addition, internet companies that act as ad publishers are reluctant to release data constrained by laws and regulations, which creates Isolated Data Island. In order to realize privacy-preserving data sharing and publishing and improve data availability as much as possible to build inventory structure, we propose a novel Local Differential Privacy (LDP) method RPM to process user impressions logs and design a privacy-preserving inventory forecasting(PPIF) framework combining clustering and sampling in this work. LDP is a promising privacy standard. There is no need for trusted third parties to participate. Each user perturbs sensitive data locally and uploads the disturbed records. The real data does not leave the local area, providing the user with the privacy preserving effect of indistinguishable individuals. There is a trade-off between privacy protection and data availability. Under availability constraints, LDP is usually used in industry only for frequency and mean statistics operations. There is plenty of room for research in designing foundational LDP technologies to support more data-mining computing tasks such as clustering. The privacy-preserving effect of RPM is demonstrated by rigorous mathematical proof, and the simulation experiments prove that the inventory estimation results obtained by using RPM are more accurate than other existing LDP methods.