Efficient land management and farming practices are critical to maintaining agricultural production, especially in Europe with limited arable land. It is very time consuming to rely on a manual field inspection of cultivated land to archive farm crops. But with the help of satellite monitoring data on the earth’s surface, it is a new vision to classify farmland based on deep learning. This article has studied the Sentinel 2 (S2) data, which are top-of-atmosphere (TOA) reflectance values at the processing level-1C (L1C) observed from some areas of Germany and France. Aiming at the problem that the interference of atmosphere and cloud coverage weakens the recognition accuracy of subsequent algorithms, a method of combining feature expansion and feature importance analysis is proposed to optimize the raw S2 data. Specifically, the new 13 spectral features are expanded based on the linear and nonlinear combination of the raw 13 spectral bands of S2. The random forest (RF) algorithm is used to score the importance of features, and the important features of each time series are selected to form a new dataset. Then, an end-to-end deep learning model has been used for training. The structure of the model is a two-layer unidirectional recurrent neural network with long short-term memory (LSTM) as the backbone. And two linear layers as the output, which form two decision-making heads, respectively, representing output classification probability and the stop decision. The results show that adding features and selecting features is beneficial for the model to improve classification accuracy and predict the classification without all of the input data. This end-to-end classification pattern with early prediction would support intelligent monitoring of farm crops with a great advantage to the implementation of various agricultural policies.

As one of the world’s major forestry countries, accurate forest-type maps in China are of great importance for the monitoring and management of forestry resources. Classifying and mapping forest types on a large scale across the country is challenging due to the complex composition of forest types, the similarity of spectral features among forest types, and the need to collect and process large amounts of data. In this study, we generated a medium-resolution (30 m) forest classification map of China using multi-source remote sensing images and local samples. A mapping framework based on Google Earth Engine (GEE) was constructed mainly using the spectral, textural, and structural features of Sentinel-1 and Sentinel-2 remote sensing images, while local acquisition data were utilized as the mapping channel for training. The proposed method includes the following steps. First, local data processing is performed to obtain training and validation samples. Second, Sentinel-1 and Sentinel-2 data are processed to improve the classification accuracy by using the enhanced vegetation index (EVI) and the red-edge position index (REPI) computed based on the S2A data. Third, to improve classification efficiency, useless bands are removed and important bands are retained through feature importance analysis. Finally, random forest (RF) is used as a classifier to train the above features, and the classification results are used for mapping and accuracy evaluation. The validation of the samples showed an accuracy of 82.37% and a Kappa value of 0.72. The results showed that the total forest area in China is 21,662,261.17 km${}^2$, of which 1,127,294.42 km${}^2$ of coniferous forests account for 52% of the total area, 981,690.98 km${}^2$ of broad-leaf forests account for 45.3 % of the total area, and 57,275.77 km${}^2$ of mixed coniferous and broad-leaf forests account for 2.6% of the total area. Upon further evaluation, we found that textural and structural features play a greater role in classification compared to spectral features. Our study shows that combining multi-source high-resolution remote sensing imagery with locally collected samples can produce forest maps for large areas. Our maps can accurately reflect the distribution of forests in China, which is conducive to forest conservation and development.

Imagining recognition of behaviors from video sequences for a machine is full of challenges but meaningful. This work aims to predict students’ behavior in an experimental class, which relies on the symmetry idea from reality to annotated reality centered on the feature space. A heteromorphic ensemble algorithm is proposed to make the obtained features more aggregated and reduce the computational burden. Namely, the deep learning models are improved to obtain feature vectors representing gestures from video frames and the classification algorithm is optimized for behavior recognition. So, the symmetric idea is realized by decomposing the task into three schemas including hand detection and cropping, hand joints feature extraction, and gesture classification. Firstly, a new detector method named YOLOv4-specific tiny detection (STD) is proposed by reconstituting the YOLOv4-tiny model, which could produce two outputs with some attention mechanism leveraging context information. Secondly, the efficient pyramid squeeze attention (EPSA) net is integrated into EvoNorm-S0 and the spatial pyramid pool (SPP) layer to obtain the hand joint position information. Lastly, the D–S theory is used to fuse two classifiers, support vector machine (SVM) and random forest (RF), to produce a mixed classifier named S–R. Eventually, the synergetic effects of our algorithm are shown by experiments on self-created datasets with a high average recognition accuracy of 89.6%.

The Named Entity Recognition (NER) task aims to locate meaningful entities in a text and classify them into pre-defined categories. NER is a fundamental task in the field of natural language processing. In practice, entities often have multiple types at different levels of granularity, and it is essential to identify various levels of categories of entities simultaneously. A deep learning-based approach treats NER as a sequence annotation problem by giving each character a unique entity label, which significantly improves the accuracy of NER. Deep learning-based NER performs well on single-category and coarse-grained NER but does not perform well on hierarchical category NER problems. This paper proposes a multi-task learning-based hierarchical category named entity recognition model to solve this problem. The model uses a multi-task learning mechanism to simultaneously learn multiple levels of information of named entity recognition tasks. Named entity recognition results at higher levels are directly constructed from lower-level recognition structures. Experiments show that both systems can significantly improve the recognition performance at each level compared to the basic multi-tasking model. The model is also compared with each level's traditional named entity recognition method. The hierarchical category named entity recognition model can sig-nificantly improve the performance of named entity recognition tasks at each level.