Nowadays, Internet of Things (IoT) gives rise to a huge amount of data. IoT nodes equipped with smart sensors can immediately extract meaningful knowledge from the data through machine learning technologies. Deep learning (DL) is constantly contributing significant progress in smart sensing due to its dramatic superiorities over traditional machine learning. The promising prospect of wide-range applications puts forwards demands on the ubiquitous deployment of DL under various contexts. As a result, performing DL on mobile or embedded platforms is becoming a common requirement. Nevertheless, a typical DL application can easily exhaust an embedded or mobile device owing to a large amount of multiply and accumulate (MAC) operations and memory access operations. Consequently, it is a challenging task to bridge the gap between deep learning and resource-limited platforms. We summarize typical applications of resource-limited deep learning and point out that deep learning is an indispensable impetus of pervasive computing. Subsequently, we explore the underlying reasons for the high computational overhead of DL through reviewing the fundamental concepts including capacity, generalization, and backpropagation of a neural network. Guided by these concepts, we investigate on principles of representative research works, as well as three types of solutions: algorithmic design, computational optimization, and hardware revolution. In pursuant to these solutions, we identify challenges to be addressed.

With the rapid development of enabling technologies like VR and AR, we human beings are on the threshold of the ubiquitous human-centric intelligence era. 6G is believed to be an indispensable cornerstone for efficient interaction between humans and computers in this promising vision. 6G is supposed to boost many human-centric applications due to its unprecedented performance improvements compared to 5G and before. However, challenges are still to be addressed, including but not limited to the following six aspects: Terahertz and millimeter-wave communication, low latency and high reliability, energy efficiency, security, efficient edge computing and heterogeneity of services. It is a daunting job to fit traditional analytical methods into these problems due to the complex architecture and highly dynamic features of ubiquitous interactive 6G systems. Fortunately, deep learning can circumvent the interpretability issue and train tremendous neural network parameters, which build mapping relationships from neural network input (status and specific requirements of a 6G application) to neural network output (settings to satisfy the requirements). Deep learning methods can be an efficient alternative to traditional analytical methods or even conquer unresolvable predicaments of analytical methods. We review representative deep learning solutions to the aforementioned six aspects separately and focus on the principles of fitting a deep learning method into specific 6G issues. Based on this review, our main contributions are highlighted as follows. (i) We investigate the representative works in a systematic view and find out some important issues like the vital role of deep reinforcement learning in the 6G context. (ii) We point out solutions to the lack of training data in 6G communication context. (iii) We reveal the relationship between traditional analytical methods and deep learning, in terms of 6G applications. (iv) We identify some frequently used efficient techniques in deep-learning-based 6G solutions. Finally, we point out open problems and future directions.

Suitable combinations of observed datasets for estimating crop model parameters can reduce the computational cost while ensuring accuracy. This study aims to explore the quantitative influence of different combinations of the observed phenological stages on estimation of cultivar-specific parameters (CPSs). We used the CROPGRO-Soybean phenological model (CSPM) as a case study in combination with the Generalized Likelihood Uncertainty Estimation (GLUE) method. Different combinations of four observed phenological stages, including initial flowering, initial pod, initial grain, and initial maturity stages for five soybean cultivars from Exp. 1 and Exp. 3 described in Table 2 are respectively used to calibrate the CSPs. The CSPM, driven by the optimized CSPs, is then evaluated against two independent phenological datasets from Exp. 2 and Exp. 4 described in Table 2. Root means square error (RMSE) (mean absolute error (MAE), coefficient of determination (R²), and Nash Sutcliffe model efficiency (NSE)) are 15.50 (14.63, 0.96, 0.42), 4.76 (3.92, 0.97, 0.95), 4.69 (3.72, 0.98, 0.95), 3.91 (3.40, 0.99, 0.96) and 12.54 (11.67, 0.95, 0.60), 5.07 (4.61, 0.98, 0.93), 4.97 (4.28, 0.97, 0.94), 4.58 (4.02, 0.98, 0.95) for using one, two, three, and four observed phenological stages in the CSPs estimation. The evaluation results suggest that RMSE and MAE decrease, and R² and NSE increase with the increase in the number of observed phenological stages used for parameter calibration. However, there is no significant reduction in the RMSEs (MAEs, NSEs) using two, three, and four observed stages. Relatively reliable optimized CSPs for CSMP are obtained by using at least two observed phenological stages balancing calibration effect and computational cost. These findings provide new insight into parameter estimation of crop models.

Motion-based hand gesture is an important scheme to allow users to invoke commands on their smartphones in an eyes-free manner. However, the existing scheme is facing some problems. On the one hand, the expression ability of one single gesture is limited. As a result, a gesture set consisting of multiple gestures is typically adopted to represent different commands. Users must memorize all gestures in order to make interaction successfully. On the other hand, the design of gestures needs to be complicated to express diverse intensions. However, complex gestures are difficult to learn and remember. In addition, complex gestures set a high recognition barrier to smart APPs. This leads to an imbalance problem. Different gestures have different recognition accuracy levels, which may result in instability of recognition precision in practical applications. To address these problems, this paper proposes a novel scheme using binary motion gestures. Only two simple gestures are required to express bit “0” and “1,” and rich information can be expressed through the permutation and combination of the two binary gestures. Firstly, four kinds of candidate binary gestures are evaluated for eyes-free interactions. Then, an online signal cutting and merging algorithm is designed to split accelerometer signals sequence into multiple separate gesture signal segments. Next, five algorithms, including Dynamic Time Warping (DTW), Naive Bayes, Decision Tree, Support Vector Machine (SVM), and Bidirectional Long Short-Term Memory (BLSTM) Network, are adopted to recognize these segments of knock gestures. The BLSTM achieves the top performance in terms of both recognition accuracy and recognition imbalance. Finally, an Android application is developed to illustrate the usability of the proposed binary gestures. As binary gestures are much simpler than traditional hand gestures, they are more efficient and user-friendly. Our scheme eliminates the imbalance problem and achieves high recognition accuracy.

Accurate prediction of crop phenological stage is essential for evaluating management strategies and assessing crop responses to environmental changes. In this work, we modified Non-dominated Sorting Genetic Algorithm with the core algorithm of PEST (MNSGA-II) and compared it to two other algorithms of Generalized Likelihood Uncertainty Estimation (GLUE) and Differential Evolution (DE) to calibrate the cultivar-specific parameters (CSPs) of CROPGRO-Soybean phenological model (CSPM) so as to exactly simulate the soybean phenology using the multi-source datasets of multi-site, multi-year, and multi-cultivar. Independent experimental data are used to validate the CSPM with the optimized parameters. The root means square error (RMSE), the mean absolute error (MAE), and coefficient of determination (R²) are used to evaluate the effects of different algorithms on calibrating the CSPs. The RMSEs (MAEs, R²) between all observed data and simulated data based on MNSGA-II, GLUE and DE are 4.28 (3.53, 0.9445) days, 4.76 (4.05, 0.9438) days and 5.17 (4.85, 0.9336) days, respectively, with little difference among the three algorithms. MNSGA-II has a certain advantage in calibration effect, and GLUE is the most stable during the repetition of each calibration. The MNSGA-II can be considered as a relatively ideal algorithm for estimating the crop model parameter. Which algorithm should be selected to calibrate the parameters of crop model according to the actual requirements. These results provide a reference to choose the suitable algorithm for estimating crop model parameter.

The virtual crop stands as a vital content in crop model research field, and has become an indispensable tool for exploring crop phenotypes. The focal objective of this undertaking is to realize three-dimensional (3D) dynamic visualization simulations of rice individual and rice populations, as well as to predict rice phenotype using virtual rice. Leveraging our laboratory’s existing research findings, we have realized 3D dynamic visualizations of rice individual and populations across various growth degree days (GDD) by integrating the synchronization relationship between the above-ground parts and the root system in rice plant. The resulting visualization effects are realistic with better predictive capability for rice morphological changes. We conducted a field experiment in Anhui Province in 2019, and obtained leaf area index data for two distinct rice cultivars at the tiller stage, jointing stage, and flowering stage. A method of segmenting leaf based on the virtual rice model is employed to predict the leaf area index. A comparative analysis between the measured and simulated leaf area index yielded relative errors spanning from 7.58% to 12.69%. Additionally, the root mean square error, the mean absolute error, and the coefficient of determination were calculated as 0.56, 0.55, and 0.86, respectively. All the evaluation criteria indicate a commendable level of accuracy. These advancements provide both technical and modeling support for the development of virtual crops and the prediction of crop phenotypes.