Small object detection is one of the difficulties in the development of computer vision, especially in the case of complex image backgrounds, and the accuracy of small object detection still needs to be improved. In this article, we present a small object detection network based on YOLOv4, which solves some obstacles that hinder the performance of traditional methods in small object detection tasks in complex road environments, such as few effective features, the influence of image noise, and occlusion by large objects, and improves the detection of small objects in complex background situations such as drone aerial survey images. The improved network architecture reduces the computation and GPU memory consumption of the network by including the cross-stage partial network (CSPNet) structure into the spatial pyramid pool (SPP) structure in the YOLOv4 network and convolutional layers after concatenation operation. Secondly, the accuracy of the model on the small object detection task is improved by adding a more suitable small object detection head and removing one used for large object detection. Then, a new branch is added to extract feature information at a shallow location in the backbone part, and the feature information extracted from this branch is fused in the neck part to enrich the small object location information extracted by the model; when fusing feature information from different levels in the backbone, the fusion weight of useful information is increased by adding a weighting mechanism to improve detection performance at each scale. Finally, a coordinated attention (CA) module is embedded at a suitable location in the neck part, which enables the model to focus on spatial location relationships and inter-channel relationships and enhances feature representation capability. The proposed model has been tested to detect 10 different target objects in aerial images from drones and five different road traffic signal signs in images taken from vehicles in a complex road environment. The detection speed of the model meets the criteria of real-time detection, the model has better performance in terms of accuracy compared to the existing state-of-the-art detection models, and the model has only 44M parameters. On the drone aerial photography dataset, the average accuracy of YOLOv4 and YOLOv5L is 42.79% and 42.10%, respectively, while our model achieves an average accuracy (mAP) of 52.76%; on the urban road traffic light dataset, the proposed model achieves an average accuracy of 96.98%, which is also better than YOLOv4 (95.32%), YOLOv5L (94.79%) and other advanced models. The current work provides an efficient method for small object detection in complex road environments, which can be extended to scenarios involving small object detection, such as drone cruising and autonomous driving.

Nowadays, tourists increasingly prefer to check the reviews of attractions before traveling to decide whether to visit them or not. To respond to the change in the way tourists choose attractions, it is important to classify the reviews of attractions with high precision. In addition, more and more tourists like to use emojis to express their satisfaction or dissatisfaction with the attractions. In this paper, we built a dataset for Chinese attraction evaluation incorporating emojis (CAEIE) and proposed an explicitly n-gram masking method to enhance the integration of coarse-grained information into a pre-training (ERNIE-Gram) and Text Graph Convolutional Network (textGCN) (E2G) model to classify the dataset with a high accuracy. The E2G preprocesses the text and feeds it to ERNIE-Gram and TextGCN. ERNIE-Gram was trained using its unique mask mechanism to obtain the final probabilities. TextGCN used the dataset to construct heterogeneous graphs with comment text and words, which were trained to obtain a representation of the document output category probabilities. The two probabilities were calculated to obtain the final results. To demonstrate the validity of the E2G model, this paper was compared with advanced models. After experiments, it was shown that E2G had a good classification effect on the CAEIE dataset, and the accuracy of classification was up to 97.37%. Furthermore, the accuracy of E2G was 1.37% and 1.35% ahead of ERNIE-Gram and TextGCN, respectively. In addition, two sets of comparison experiments were conducted to verify the performance of TextGCN and TextGAT on the CAEIE dataset. The final results showed that ERNIE and ERNIE-Gram combined TextGCN and TextGAT, respectively, and TextGCN performed 1.6% and 2.15% ahead. This paper compared the effects of eight activation functions on the second layer of the TextGCN and the activation-function-rectified linear unit 6 (RELU6) with the best results based on experiments.

With the continuous development of the internet and big data, modernization and informatization are rapidly being realized in the agricultural field. In this line, the volume of agricultural news is also increasing. This explosion of agricultural news has made accurate access to agricultural news difficult, and the spread of news about some agricultural technologies has slowed down, resulting in certain hindrance to the development of agriculture. To address this problem, we apply NLP to agricultural news texts to classify the agricultural news, in order to ultimately improve the efficiency of agricultural news dissemination. We propose a classification model based on ERNIE + DPCNN, ERNIE, EGC, and Word2Vec + TextCNN as sub-models for Chinese short-agriculture text classification (E3W), utilizing the GreedySoup weighting strategy and multi-model combination; specifically, E3W consists of four sub-models, the output of which is processed using the GreedySoup weighting strategy. In the E3W model, we divide the classification process into two steps: in the first step, the text is passed through the four independent sub-models to obtain an initial classification result given by each sub-model; in the second step, the model considers the relationship between the initial classification result and the sub-models, and assigns weights to this initial classification result. The final category with the highest weight is used as the output of E3W. To fully evaluate the effectiveness of the E3W model, the accuracy, precision, recall, and F1-score are used as evaluation metrics in this paper. We conduct multiple sets of comparative experiments on a self-constructed agricultural data set, comparing E3W and its sub-models, as well as performing ablation experiments. The results demonstrate that the E3W model can improve the average accuracy by 1.02%, the average precision by 1.62%, the average recall by 1.21%, and the average F1-score by 1.02%. Overall, E3W can achieve state-of-the-art performance in Chinese agricultural news classification.

<abstract> <p>The automatic text summarization task faces great challenges. The main issue in the area is to identify the most informative segments in the input text. Establishing an effective evaluation mechanism has also been identified as a major challenge in the area. Currently, the mainstream solution is to use deep learning for training. However, a serious exposure bias in training prevents them from achieving better results. Therefore, this paper introduces an extractive text summarization model based on a graph matrix and advantage actor-critic (GA2C) method. The articles were pre-processed to generate a graph matrix. Based on the states provided by the graph matrix, the decision-making network made decisions and sent the results to the evaluation network for scoring. The evaluation network got the decision results of the decision-making network and then scored them. The decision-making network modified the probability of the action based on the scores of the evaluation network. Specifically, compared with the baseline reinforcement learning-based extractive summarization (Refresh) model, experimental results on the CNN/Daily Mail dataset showed that the GA2C model led on Rouge-1, Rouge-2 and Rouge-A by 0.70, 9.01 and 2.73, respectively. Moreover, we conducted multiple ablation experiments to verify the GA2C model from different perspectives. Different activation functions and evaluation networks were used in the GA2C model to obtain the best activation function and evaluation network. Two different reward functions (Set fixed reward value for accumulation (ADD), Rouge) and two different similarity matrices (cosine, Jaccard) were combined for the experiments.</p> </abstract>