Existing single-stage target detection algorithms are insensitive to nodule detection in lung nodule detection， multiple up-samplings during feature extraction by Convolutional Neural Network （CNN） has difficult feature extraction and poor detection effect， and the existing pulmonary nodule detection algorithm models are complex and not conductive to practical application employment and implementation. To address the above problems， a real-time pulmonary nodule detection algorithm combining attention mechanism and multipath fusion was proposed， based on which the up-sampling algorithm was improved to effectively increase the detection accuracy of lung nodules and speed of model inference， the model size was small and easy to deploy. Firstly， the hybrid attention mechanism of channel and space was fused in the backbone network part of feature extraction. Secondly， the sampling algorithm was improved to enhance the quality of generated feature maps. Finally， the channels were established between different paths in the enhanced feature extraction network part to achieve the fusion of deep and shallow features， so the semantic and location information at different scales was fused. Experimental results on LUNA16 dataset show that， compared to the original YOLOv5s algorithm， the proposed algorithm achieves an improvement of 9.5， 6.9， and 8.7 percentage points in precision， recall， and average precision， respectively， with a frame rate of 131.6 frames/s， and a model weight file of only 14.2 MB， demonstrating that the proposed algorithm can detect lung nodules in real time with much higher accuracy than existing single-stage detection algorithms such as YOLOv3 and YOLOv8.
Deep reinforcement learning algorithms are difficult to learn optimal policy through interaction with environment in reward sparsity environments， so that the intrinsic reward needs to be built to guide the update of algorithms. However， there are still some problems in this way： 1） statistical inaccuracy of state classification will misjudge reward value， thereby causing the agent to learn wrong behavior； 2） due to the strong ability of the prediction network to identify state information， the state freshness generated by the intrinsic reward decreases， which affects the learning effect of the optimal policy； 3） due to the random state transition， the information of the teacher strategies is not effectively utilized， which reduces the agent’s ability to explore the environment. To solve the above problems， a reward construction mechanism combining prediction error of stochastic generative network with hash discretization statistics， namely RGNP-HCE （Randomly Generated Network Prediction and Hash Count Exploration）， was proposed， and the knowledge of multi-teacher policy was transferred to student policy through distillation. In RGNP-HCE mechanism， the fusion reward was constructed through the idea of curiosity classification. In specific， the global curiosity reward was constructed by stochastic generative network’s prediction error between multiple episodes， and the local curiosity reward was constructed by hash discretization statistics in one episode， which guaranteed the rationality of intrinsic rewards and the correctness of policy gradient updates. In addition， multi-teacher policy distillation provides students with multiple reference directions for exploration， which improved environmental exploration ability of the student policy effectively. Finally， in the test environments of Montezuma’s Revenge and Breakout， experiment of comparing the proposed mechanism with four current mainstream deep reinforcement learning algorithms was carried out， and policy distillation was performed. The results show that compared with the average performance of current high-performance deep reinforcement learning algorithms， the average performance of RGNP-HCE mechanism in both test environments is improved， and the distilled student policy is further improved in average performance， indicating that RGNP-HCE mechanism and policy distillation are effective in improving the exploration ability of agent.
In order to construct a transparent and trustworthy recommendation mechanism， relevant research works mainly provide reasonable explanations for personalized recommendation through explainable recommendation mechanisms. However， there are three major limitations of the existing explainable recommendation mechanism： 1） using correlations only can provide rational explanations rather than causal explanations， and using paths to provide explanations will bring privacy leakage； 2） the problem of sparse user feedback is ignored， so it is difficult to guarantee the fidelity of explanations； 3） the granularities of explanations are relatively coarse， and users’ personalized preferences are not considered. To solve the above problems， an explainable recommendation mechanism ERCKCI based on Collaborative Knowledge Graph （CKG） and counterfactual inference was proposed. Firstly， based on the user’s own behavior sequence， the counterfactual inference was used to achieve high-sparsity causal decorrelation by using the casual relations， and the counterfactual explanations were derived iteratively. Secondly， in order to improve the fidelity of explanations， not only the CKG and the neighborhood propagation mechanism of the Graph Neural Network （GNN） were used to learn users’ and items’ representations based on single time slice； but also the user long-short term preference were captured to enhance user preference representation through self-attention mechanism on multiple time slices. Finally， via a higher-order connected subgraph of the counterfactual set， the multi-granularity personalized preferences of user was captured to enhance counterfactual explanations. To verify the effectiveness of ERCKCI mechanism， comparison experiments were performed on the public datasets MovieLens（100k）， Book-crossing and MovieLens（1M）. The obtained results show that compared with the Explainable recommendation based on Counterfactual Inference （ECI） algorithm under the Relational Collaborative Filtering （RCF） recommendation model on the first two datasets， the proposed mechanism has the explanation fidelity improved by 4.89 and 3.38 percentage points respectively， the size of CF set reduced by 63.26% and 66.24% respectively， and the sparsity index improved by 1.10 and 1.66 percentage points respectively； so the explainability is improved effectively by the proposed mechanism.
Aiming at the problems of low detection precision， poor robustness， and imperfect related systems in the current small object detection of electric vehicle helmet， an electric vehicle helmet detection model was proposed based on improved YOLOv5s algorithm. In the proposed model， Convolutional Block Attention Module （CBAM） and Coordinate Attention （CA） module were introduced， and the improved Non-Maximum Suppression （NMS） - Distance Intersection over Union-Non Maximum Suppression （DIoU-NMS） was used. At the same time， multi-scale feature fusion detection was added and densely connected network was combined to improve feature extraction effect. Finally， a helmet detection system for electric vehicle drivers was established. The improved YOLOv5s algorithm had the mean Average Precision （mAP） increased by 7.1 percentage points when the Intersection over Union （IoU） is 0.5， and Recall increased by 1.6 percentage points compared with the original YOLOv5s on the self-built electric vehicle helmet wearing dataset. Experimental results show that the improved YOLOv5s algorithm can better meet the requirements for detection precision of electric vehicles and the helmets of their drivers in actual situations， and reduce the incidence rate of electric vehicle traffic accidents to a certain extent.
Recent deep learning achieves great attention on the tasks of Drug-Target Affinity （DTA）. However， most existing works embed the molecular single structure as a vector， while ignoring the information gain provided by multi-aspect fusion of molecular features to the final feature representation. To address the feature incompleteness problem of single-structured molecules， an end-to-end deep learning method based on attentive fusion of multi-aspect molecular features was proposed for DTA prediction. Multi-aspect molecular structure embedding （Mas） and Multi-attention feature fusion （Mat） are the core modules of the proposed method. Firstly， the multi-aspect molecular structure was embedded into the feature vector space by Mas module. Secondly， the attention mechanism of molecular feature level was incorporated for the weighted fusion of molecular features from different aspects through Mat module. Thirdly， feature cascade of the above two was performed according to Drug-Target Interaction （DTI）. Finally， the fully connected neural network was used to realize the regression prediction of the affinity. Experiments on Davis and KIBA datasets were carried out to evaluate the influence of training ratio， multi-aspect feature incorporation， multi-attention fusion， and related parameters on the performance of affinity prediction. Compared with the GraphDTA method， the proposed method has the Mean Square Error （MSE） reduced by 4.8% and 6% on the two datasets， respectively. Experimental results show that attentive fusion of multi-aspect molecular features can capture the molecular features that are more relevant for linkages on protein targets.
