
Automatic detection of videos with outstanding situations is a practical issue that needs to be studied in many events of different fields with common length and frequency of occurrence for instance: meetings, musicals, sports events that the user uploads regularly, one of the concerned areas is the highlights in football videos. The matches of the annual top leagues and between nations within federations form a huge database in need of different purposes in which requires the specific model assisting in extracting outstanding situations. Besides, building a reliable and accurate model requires an appropriate approach, a large amount of training data (diverse, accurate, clear data), which need to be assigned label correspondingly. The Convolutional Neural Network (CNN) was chosen as an approach to help building a smart system as a foundation, combined with a proposed new method for synthesizing results called the adaptive threshold towards specific data, along with the optimization model to draw a reliable conclusion. The work was proceeded on a video data set of the top four teams of the English Premier League (ELP) 2018–2019 and a randomly selected dataset on the Internet.
Geostatistics is a branch of statistics that focuses on spatial data sets. Geostatistics is used for prediction in many fields such as ore mining, petroleum geology, climate, geography, environment. CoKriging can be seen as a point interpolation, which requires a point map as input and which returns a raster map with estimations and optionally an error map. CoKring is a prediction method for problems with two or more parameters, including one main parameter. Tthe costs of installing new observational stations to observe metropolitan air pollution sources, as PM10 (Particulate Matter), CO (Cacbon monoxit), SO2 (Dioxit Sunfua) and NO2 (oxit Nito) concentrations are high. In this study, analysis of air pollution of 16 stations monitored in 2018 year was carried out. Geostatistics have been used by many researchers to effectively predict air pollution, ore mining, and groundwater levels.
In this paper, we survey hardware/software co-design approaches in the literature to accelerate Convolutional neural networks, one of the two successful forms of Deep Neural Networks. We classify these approaches according to target platforms used to accelerate CNNs, including FPGA-based and ASIC-based. Due to the flexibility of FPGAs, we mainly focus on FPGA-based designs. These designs are categorized into sub-classes according to optimization techniques used. We then analyze in detail to compare FPGA-based hardware accelerator systems for CNNs regarding working frequency and performance efficiency. Through the survey, we also identify open issues for this research topic. These challenges can be research directions for optimizing performance, accuracy, and energy consumption of hardware accelerator systems for CNN using the hardware/software co-design approach.
Plant diseases is one of the most influential factors in agricultural production. It can affect product quality, quantity, or yield of crops. Diagnosis of plant diseases is made mainly based on the experience of farmers. This work is done based on the naked eye. It is often misleading, time-consuming, and laborious. Machine learning methods based on leaf images have been proposed to improve disease identification. Transfer learning is accepted and proven to be effective. In this paper, we used the transfer learning method to classify apple tree diseases. The research data were used from the Fine-Grained Visual Categorization (FGVC7) Kaggle PLANT PATHOLOGY 2020, expert-annotated to create a pilot dataset for apple scab, cedar apple rust, multiple diseases, and healthy leaves. The InceptionV3 architecture trained with the Adam optimizer attained the highest validation accuracy.
The context-aware computing paradigm introduces environments, known as smart spaces, which can unobtrusively and proactively assist their users. These systems are currently mostly implemented on mobile platforms considering various techniques, including ontology-driven multi-agent rule-based reasoning. Rule-based reasoning is a relatively simple model that can be adapted to different real-world problems. It can be developed considering a set of assertions, which collectively constitute the working memory, and a set of rules that specify how to act on the assertion set. However, the size of the working memory is crucial when developing context-aware systems in resource constrained devices such as smartphones and wearables. In this paper, we discuss rule-based context-aware systems and techniques for determining the required working memory size for a fixed set of rules.