The governance of rural living environments is one of the important tasks in the implementation of a rural revitalization strategy. At present, the illegal behaviors of random construction and random storage in public spaces have seriously affected the effectiveness of the governance of rural living environments. The current supervision on such problems mainly relies on manual inspection. Due to the large number and wide distribution of rural areas to be inspected, this method is limited by obvious disadvantages, such as low detection efficiency, long-time spending, and huge consumption of human resources, so it is difficult to meet the requirements of efficient and accurate inspection. In response to the difficulties encountered, a low-altitude remote sensing inspection method on rural living environments was proposed based on a modified YOLOv5s-ViT (YOLOv5s-Vision Transformer) in this paper. First, the BottleNeck structure was modified to enhance the multi-scale feature capture capability of the model. Then, the SimAM attention mechanism module was embedded to intensify the model’s attention to key features without increasing the number of parameters. Finally, the Vision Transformer component was incorporated to improve the model’s ability to perceive global features in the image. The testing results of the established model showed that, compared with the original YOLOv5 network, the Precision, Recall, and mAP of the modified YOLOv5s-ViT model improved by 2.2%, 11.5%, and 6.5%, respectively; the total number of parameters was reduced by 68.4%; and the computation volume was reduced by 83.3%. Relative to other mainstream detection models, YOLOv5s-ViT achieved a good balance between detection performance and model complexity. This study provides new ideas for improving the digital capability of the governance of rural living environments.

The plant disease recognition model based on deep learning has shown good performance potential. However, high complexity and nonlinearity lead to the low transparency and poor interpretability of such models. These limitations greatly limit the deployment and application of such models in field scenarios. To solve the above problems, we propose a dense caption generative model, Veg DenseCap. This model takes vegetable leaf images as input and uses object detection technology to locate abnormal parts of the leaf and identify the disease results. More importantly, it can describe the disease features it sees in natural language, and users can judge whether the relevant features are semantically consistent with human cognition based on these descriptions. First of all, a dataset containing Chinese feature description statements for images of 10 leaf diseases involving two vegetables (cucumber and tomato) was established. Secondly, Faster R-CNN was used as a disease detector to extract visual features of diseases, and LSTM was used as a language generator to generate description statements for disease features. Finally, the Convolutional Block Attention Module (CBAM) and the Focal Loss function were employed to overcome the imbalance between positive and negative samples and the weak performance of Faster R-CNN in obtaining key features. According to the test results, the Intersection-over-Union (IoU) and Meteor joint evaluation index of Veg-DenseCap achieved a mean Average Precision (mAP) of 88.0% on the dense captioning dataset of vegetable leaf disease images, which is 9.1% higher than that of the classical FCLN model. The automatically generated description statements are characterized by advantages of accurate feature description, correct grammar, and high diversity.