Search

This survey provides an overview of recent advancements in wireless soil sensing technologies, including the sensors and communication systems used to manage and enhance soil health and crop production for sustainable agricultural practices. The surveyed technologies are categorized by the target sensing parameters, such as moisture, temperature, salinity, pH, nutrient concentration, pressure, and heavy metal content. Special focus is placed on the passive and semi-passive sensing principles for soil data collection, communication systems for transmitting the sensed data, and machine learning approaches for data processing and analysis to assist in informed decision-making in the realization of sustainable agriculture. By comparing existing technologies, we highlight the advantages and disadvantages of these sensing mechanisms, summarize existing challenges of the surveyed soil sensing systems, and identify future research directions that need attention.

Recently, a cellular automata learning and prediction (CALP) model was proposed that considered images as living cells and used cellular automata to evolve and synthetically augment handwritten data. In this paper, another data augmentation method is proposed inspired by the flocking pattern of birds. It is proposed that any image sample in a handwritten data set can be represented as an assembly of birds. Each bird is specifically located at a point or pixel to collectively form an image. Using a well-defined flocking mechanism, the new positions for these birds can be calculated. Each snapshot of the new position can be considered as a new version of the original image, thus generating more data. The best flocking pattern is determined using biologically-inspired genetic algorithms. The quality of the synthetic data is demonstrated by using it in a handwritten pattern recognition system. For this purpose, the CALP framework is used with few modifications. It is shown that using the new data along with the original data to train classifiers improves the accuracy of classifiers than when they are trained using only the original data. Following the same test bed as CALP, 25 different data sets and classifier pairs were used for experimentation. The experimental results show that by also using the data set generated through flocking, the performance of the base classifiers is significantly improved even when smaller training data is used. We also compare the performance of the proposed technique with other state-of-the-art oversampling methods.

Deep learning methods have achieved state-of-the-art performance on synthetic aperture radar (SAR) target recognition tasks in recent years. However, obtaining sufficient SAR images for training these deep learning methods is costly in time and labor. This paper focuses on recognizing targets with a few training samples, that is, few-shot target recognition. We combine deep neural networks’ powerful feature representation capabilities with the nonparametric flexibility of Gaussian processes (GPs) and propose a few-shot recognition model based on deep kernel learning. Deep neural networks map input samples into a low-dimensional embedding space. GPs employ a family of kernel functions to measure the similarity between embedded samples and classify them. During training, the model builds diverse related tasks to learn kernel functions with parameters shared across few-shot tasks. These learned kernel functions define common prior knowledge that can be transferred to unseen tasks. During testing, the model can recognize novel tasks with a few samples based on learned kernel functions. We conducted extensive experiments on a widely-used real SAR dataset to evaluate the model’s effectiveness. The test results demonstrate that our model is superior to several recently proposed few-shot recognition methods.

Aerospace control software is the most important part of aerospace software. Since its potential defects endanger life and safety, there are strict requirements on product quality. Therefore, efficient and reliable software testing is essential. The traditional testing method has been challenging to meet its development requirements, and software automation testing has gradually become the main tool for testing aerospace control software. For the automation testing of aerospace control software, the core problem is to locate the GUI widgets on the software screenshots and identify their intent, which directly affects the accuracy of the test. Because of this, we use the widget recognition technology based on image matching and use the image understanding and analysis technology to extract the widget image in the screenshots. After obtaining the widget image, we use a convolutional neural network to extract image features and use the encoder module to encode the extracted information features as a tensor. The decoder module generates a word sequence conditional on tensor and previous output based on the encoded information. We also conduct an empirical study to evaluate the accuracy of widget recognition and intention generation. For widget recognition, our average IoU reached 0.81. For widget intent generation, our model BLEU-1 is 0.567, BLEU-2 is 0.356, BLEU-3 is 0.261, BLEU-4 is 0.131. The results show that our method is very effective.