Nighttime object detection faces numerous challenges compared to daytime object detection， due to low-light conditions and the scarcity of high-quality labeled data， which hinder feature extraction and degrade detection accuracy. Therefore， a domain-adaptive object detection method for nighttime images was proposed. Firstly， a nighttime domain-adaptive photometric alignment module was designed to convert a labeled daytime source domain image into a corresponding nighttime target domain image， that is bridging the gap between source and target domains through photometric alignment， thereby solving the problem of difficulty in obtaining accurate nighttime object labels under low-light conditions. Secondly， a hybrid CNN-Transformer model was used as a detector， in which using CSwin Transformer was used as a backbone network to extract multi-level image features and these features were input into a feature pyramid network， thus enhancing the model's capability for multi-scale object detection. Finally， the Outlook attention module was introduced to address the loss of image details caused by insufficient lighting， thereby enhancing the model's robustness under varying lighting conditions， shadows， and other complex environments. Experimental results demonstrate that the proposed method achieves the mean Average Precision （mAP） @0.5 of 50.0% on the public dataset BDD100K， which is improved by 4.2 percentage points compared to 2PCNet （two-Phase Consistency Network） method； and the mAP@0.5 reached 45.4% on the public dataset SODA10M， which is improved by 0.9 percentage points compared to SFA （Sequence Feature Alignment） method.

Multi-round Conversational Recommendation System （CRS） obtains real-time information of users interactively， thus performing better than traditional recommendation methods such as collaborative filtering based method. However， existing CRS suffers from problems inaccurate mining of user preferences， too many conversational rounds required and inappropriate recommendation moments. Aiming at these problems， a new conversational recommendation algorithm based on deep reinforcement learning considering user’s multi-granularity feedback information was proposed. Different from existing CRS， in each conversation， the feedback of users on items themselves and more fine-grained item attributes was considered by the proposed algorithm at the same time. Then， users， items and attribute features of items were updated online by using the collected multi-granularity feedback， and the environment state after each round of conversation was analyzed by Deep Q-Network （DQN） algorithm. As a result， more appropriate and reasonable decisions were made by the system， and the reasons of why user buying items were analyzed and the users’ real-time preferences were mined comprehensively with fewer conversation rounds. Experimental results on two real datasets show that compared with Simple Conversational Path Reasoning （SCPR） algorithm， the proposed algorithm has the 15 turns success rate increased by 46.5%， and the 15 average turns decreased by 0.314 rounds in Last.fm dataset， while it maintains the same level of success rate but the 15 average turns decreased by 0.51 rounds in Yelp dataset.

In order to quickly and effectively reduce digital image from multi-wavelet transform domain to space domain and get the reduced image with better visual result, a reduction method including inverse transform and post filtering was proposed. The proposed method decomposed image from space domain to transform domain by pre-filtering and multi-wavelet transform, and then recombined high and low frequency components, finally reduced spatial domain image without interpolation zero. The experimental results show that compared with the original image, the error value of more than 90% pixels of restored image is less than 0.0001.

Most existing steganography algorithms for color images are designed by directly applying steganography for gray-scale images to color channel without employing color visual perception mechanism. A large-capacity blind steganography algorithm for color images based on visual perception mechanism in YUV space was proposed. The main idea is to analyze the local complexity of the carrier image via standard deviation on the Y component, split the message into two parts using the color visual perception mechanism, and then hide them in Y and V components respectively. Being less flexible, the U component was simply used as an information indicator. A large number of experiments were carried out showing that the proposed algorithm still maintains good imperceptibility compared with other methods after embedding large-capacity information, and can resist both the histogram contrast and RS steganalysis. Thus, the proposed algorithm is reasonable, easy to implement, and effective.

The control and treatment of water pollution is an important issue in China. To meet the lack of remote monitoring of water, the ship video transmission and protection system based on 3G network was proposed. The structure of the system was described, and the characteristics of the 3G network video transmission were analyzed. The achievement of smooth real-time video transmission was based on 3G network, simple reliable user datagram protocol, H.264 video codec, and Quality of Service (QoS) control. The results show that the system is effective, and it can be applied to real-time video surveillance of water.