Search

In recent years, researchers have explored reinforcement learning based object detection methods. However, existing methods always suffer from barely satisfactory performance. The main reasons are that current reinforcement learning based methods generate a sequence of inaccurate regions without a reasonable reward function, and regard the non-optimal one at the final step as the detection result for lack of an effective region selection and refinement strategy. To tackle the above problems, we propose a novel reinforcement learning based object detection framework, namely ReinforceNet, possessing the capability of the region selection and refinement by integrating reinforcement learning agents’ action space with Convolutional Neural Network based feature space. In ReinforceNet, we redevelop a reward function that enables the agent to be trained effectively and provide more accurate region proposals. In order to further optimize them, we design Convolutional Neural Network based region selection network (RS-net) and bounding box refinement network (BBR-net). Particularly, the former consists of two sub-networks: Intersection-of-Union network (IoU-net) and Completeness network (CPL-net) which are employed jointly for selecting the optimal region proposal. The latter aims to refine the selected one as the final result. Extensive experimental results on two standard datasets PASCAL VOC 2007 and VOC 2012 demonstrate that ReinforceNet is capable of improving the region selection and learning better agent action representations for reinforcement learning, resulting in the state-of-the-art performance.

Specialized control of pests and diseases have been\ud a high-priority issue for agriculture industry in many countries.\ud On account of automation and cost-effectiveness, image analytic based pest recognition systems are widely utilized in practical crops prevention applications. But due to powerless handcrafted features, current image analytic approaches achieve low accuracy and poor robustness in practical large-scale multi-class pest detection and recognition. To tackle this problem, this paper proposes a novel deep learning based automatic approach using hybrid and local activated features for pest monitoring solution. In the presented method, we exploit the global information from feature maps to build our Global activated Feature Pyramid Network (GaFPN) to extract pests’ highly discriminative features across various scales over both depth and position levels. It makes changes of depth or spatial sensitive features in pest images more visible during downsampling. Next, an improved pest localization module named Local activated Region Proposal Network (LaRPN) is proposed to find the precise pest objects’ positions by augmenting contextualized and attentional information for\ud feature completion and enhancement in local level. The approach is evaluated on our 7-year large-scale pest dataset containing 88.6K images (16 types of pests) with 582.1K manually labelled pest objects. The experimental results show that our solution performs over 75.03% mAP in industrial circumstances, which outweighs two other state-of-the-art methods: Faster R-CNN with mAP up to 70% and FPN mAP up to 72%. Our code and dataset will be made publicly available.

Pests cause significant damage to agricultural crops and reduce crop yields. Use of manual methods of pest forecasting for integrated pest management is labor-intensive and time-consuming. Here, we present an automatic system for monitoring pests in large fields, with the aim of replacing manual forecasting. The system comprises an automatic detection and counting system and a human-computer data statistical fitting system. Image data sets of the target pests from large fields are first input into the system. The number of pests in the image is then counted both manually and using the automatic system. Finally, a mapping relationship between counts obtained using the automated system and by agricultural experts is established using the statistical fitting system.Trends in the pest-count curves produced using the manual and automated counting methods were very similar. To sample the number of pests for manual statistics, plants were shaken to transfer the pests from the plant to a plate. Hence, pests hiding within plant crevices were also sampled and included in the count, whereas the automatic method counted only the pests visible in the images. Therefore, the computer index threshold was much lower than the manual index threshold. However, the proposed system correctly reflected trends in pest numbers obtained using computer vision.The experimental results demonstrate that our automatic pest-monitoring system can generate pest grades and can replace manual forecasting methods in large fields. © 2021 Society of Chemical Industry.

The recent explosion of large volume of standard dataset of annotated images has offered promising opportunities for deep learning techniques in effective and efficient object detection applications. However, due to a huge difference of quality between these standardized dataset and practical raw data, it is still a critical problem on how to maximize utilization of deep learning techniques in practical agriculture applications. Here, we introduce a domain-specific benchmark dataset, called AgriPest, in tiny wild pest recognition and detection, providing the researchers and communities with a standard large-scale dataset of practically wild pest images and annotations, as well as evaluation procedures. During the past seven years, AgriPest captures 49.7K images of four crops containing 14 species of pests by our designed image collection equipment in the field environment. All of the images are manually annotated by agricultural experts with up to 264.7K bounding boxes of locating pests. This paper also offers a detailed analysis of AgriPest where the validation set is split into four types of scenes that are common in practical pest monitoring applications. We explore and evaluate the performance of state-of-the-art deep learning techniques over AgriPest. We believe that the scale, accuracy, and diversity of AgriPest can offer great opportunities to researchers in computer vision as well as pest monitoring applications.

Question classification is a basic work in natural language processing, which has an important influence on question answering. Due to question sentences are complicated in many specific domains contain a large number of exclusive vocabulary, question classification becomes more difficult in these fields. To address the specific challenge, in this paper, we propose a novel hierarchical hybrid deep network for question classification. Specifically, we first take advantages of word2vec and a synonym dictionary to learn the distributed representations of words. Then, we exploit bi-directional long short-term memory networks to obtain the latent semantic representations of question sentences. Finally, we utilize convolutional neural networks to extract question sentence features and obtain the classification results by a fully-connected network. Besides, at the beginning of the model, we leverage the self-attention layer to capture more useful features between words, such as potential relationships, etc. Experimental results show that our model outperforms common classifiers such as SVM and CNN. Our approach achieves up to 9.37% average accuracy improvements over baseline method across our agricultural dataset.

Effective pest management and control are the key factors in the agricultural food safety field. Therefore, the automatic monitoring and precise recognition of crop pests have a high practical value in the process of agricultural planting. Over the years, pest recognition and detection results have been rapidly improved with the development of deep learning-based methods. Although promising, these methods still have limited efficiency and precision to detect crop pests with very small scales, deteriorating their effectiveness. The main reason is that current deep learning-based methods may not be able to extract sufficient detailed appearance features for small pest objects in an image, making it difficult to train a classifier to detect and distinguish small pests from the backgrounds or similar objects. To address the small pest recognition and detection problem, in this paper, we instead seek to recast the current region proposal network and perform more details in different scales for easier small pest detection. Inspired by the visual attention system, we first introduce attention mechanism into the Residual network for obtaining richer pest feature appearance, especially the detailed features of small object pests; Then, to make the region proposal network (RPN) obtain more high-quality object proposals for easier detection, a sampling-balanced region proposal generation network is proposed for improving pest detection accuracy. Furthermore, we devise a novel adaptive region of interest (RoI) selection method to learn features from different levels of the feature pyramid. Several experiments were conducted on the proposed AgriPest21 dataset, and our method can achieve an average recall of 89.0% and mAP of 78.7%, outperforming other state-of-the-art methods, including SSD, RetinaNet, Free-Anchor, PISA, Grid RCNN, and Cascade RCNN detectors.

Monitoring pest in agriculture has been a high-priority issue all over the world. Computer vision techniques are widely utilized in practical crop pest prevention applications due to the rapid development of artificial intelligence technology. However, current deep learning image analytic approaches achieve low accuracy and poor robustness in agriculture pest monitoring task. This paper targets at this challenge by proposing a novel two-stage deep learning based automatic pest monitoring system with hybrid global and local activated feature. In this approach, a Global activated Feature Pyramid Network (GaFPN) is firstly proposed for extracting highly representative features of pests over both depth and spatial position activation levels. Then, an improved Local activated Region Proposal Network (LaRPN) augmenting contextual and attentional information is represented for precisely locating pest objects. Finally, we design a fully connected neural network to estimate the severity of input image under the detected pests. The experimental results on our 88.6K images dataset (with 16 types of common pests) show that our approach outweighs the state-of-the-art methods in industrial circumstances.

Neural Machine Translation (NMT) has recently achieved the state-of-the-art in many machine translation tasks, but one of the challenges that NMT faces is the lack of parallel corpora, especially for low-resource language pairs. And the result is that the performance of NMT is much less effective for low-resource languages. To address this specific problem, in this paper, we describe a novel NMT model that is based on encoder-decoder architecture and relies on character-level inputs. Our proposed model employs Convolutional Neural Networks (CNN) and highway networks over character inputs, whose outputs are given to an encoder-decoder neural machine translation network. Besides, we also present two other approaches to improve the performance of the low-resource NMT system much further. First, we use language modeling implemented by denoising autoencoding to pre-train and initialize the full model. Second, we share the weights of the front few layers of two encoders between two languages to strengthen the encoding ability of the model. We demonstrate our model on two low-resource language pairs. On the IWSLT2015 English-Vietnamese translation task, our proposed model obtains improvements up to 2.5 BLEU points compared to the baseline. We also outperform the baseline approach more than 3 BLEU points on the CWMT2018 Chinese-Mongolian translation task.

Automatic crop disease recognition in the wild is a challenging topic in modern intelligent agriculture due to the appearance variances and cluttered background among crop diseases. To overcome these obstacles, the popular methods are to design a Convolutional Neural Network (CNN) model that extracts visual features and identifies crop disease images based on these features. These methods work well on laboratory environment under simple background but achieve low accuracy and poor robustness in processing the raw images captured from practical fields that contain inevitable noises. In this case, Internet of Things (IoT) is attracting increasing attention, with many alternatives to collect high-level contextual information that helps modern recognition system to effectively identify crop diseases in the wild. Motivated by the usefulness of agricultural IoT, a deep learning system using a novel approach named Multi-Context Fusion Network (MCFN), is developed to be deployed in agricultural IoT towards practical crop disease recognition in the wild. Our MCFN firstly adopts a standard CNN backbone to extract highly discriminative and robust visual features from over 50,000 in-field crop disease samples. Next, we exploit contextual features collected from image acquisition sensors as prior information to assist crop disease classification and reduce false positives in our presented ContextNet. Finally, a deep fully connected network is designed to fuse visual features as well as contextual features and output the crop disease prediction. Experimental results on 77 common crop diseases captured in our newly built domain specific dataset show that MCFN with the deep fusion model outperforms the state-of-the-art methods in wild crop disease recognition, and achieves a good identification accuracy of 97.5%.

Automatic in-field pest detection and recognition using mobile vision technique is a hot topic in modern intelligent agriculture, but suffers from serious challenges including complexity of wild environment, detection of tiny size pest and classification of multiple classes of pests. While recent deep learning based mobile vision techniques have shown some success in overcoming above issues, one key problem is that towards large-scale multiple species of pest data, imbalanced classes significantly reduce their detection and recognition accuracy. In this paper, we propose a novel two-stages mobile vision based cascading pest detection approach (DeepPest) towards large-scale multiple species of pest data. This approach firstly extracts multi-scale contextual information of the images as prior knowledge to build up a context-aware attention network for initial classification of pest images into crop categories. Then, a multi-projection pest detection model (MDM) is proposed and trained by crop-related pest images. The role of MDM can combine pest contextual information from low-level convolutional layers with these in high-level convolutional layers for generating the super-resolved feature. Finally, we utilize the attention mechanism and data augmentation to improve the effectiveness of in-field pest detection. We evaluate our method on our newly established large-scale dataset In-Field Pest in Food Crop (IPFC) and sufficient experimental results show that DeepPest proposed in this paper outperforms state-of-the-art object detection methods in detecting in-field pest.

A knowledge model for knowledge base system in the Web environment is proposed, which includes problem description layer and knowledge layer. A knowledge base is made up of a set of knowledge models. Knowledge models are used to support knowledge representation and reasoning. Visual knowledge modeling tool and visual knowledge service tool based on Web environment realize the construction of knowledge base system in Web environment. The knowledge base system supports the construction of agricultural expert system. By comparison with CommonKADS, it is proved that this method can improve the efficiency of knowledge acquisition, management and maintenance in Web environment. This method also has the reference significance to the construction of the knowledge base system in the Web environment in other domains.

More web pages are widely applying AJAX (Asynchronous JavaScript XML) due to the rich interactivity and incremental communication. By observing, it is found that the AJAX contents, which could not be seen by traditional crawler, are well-structured and belong to one specific domain generally. Extracting the structured data from AJAX contents and annotating its semantic are very significant for further applications. In this paper, a structured AJAX data extraction method for agricultural domain based on agricultural ontology was proposed. Firstly, Crawljax, an open AJAX crawling tool, was overridden to explore and retrieve the AJAX contents; secondly, the retrieved contents were partitioned into items and then classified by combining with agricultural ontology. HTML tags and punctuations were used to segment the retrieved contents into entity items. Finally, the entity items were clustered and the semantic annotation was assigned to clustering results according to agricultural ontology. By experimental evaluation, the proposed approach was proved effectively in resource exploring, entity extraction, and semantic annotation.

Feature optimization is important to agricultural text mining. Usually, the vector space model is used to represent text documents. However, this basic approach still suffers from two drawbacks: the curse of dimension and the lack of semantic information. In this paper, a novel ontology-based feature optimization method for agricultural text was proposed. First, terms of vector space model were mapped into concepts of agricultural ontology, which concept frequency weights are computed statistically by term frequency weights; second, weights of concept similarity were assigned to the concept features according to the structure of the agricultural ontology. By combining feature frequency weights and feature similarity weights based on the agricultural ontology, the dimensionality of feature space can be reduced drastically. Moreover, the semantic information can be incorporated into this method. The results showed that this method yields a significant improvement on agricultural text clustering by the feature optimization.

Although the genomics data are accumulated in an exponential growth, the molecular complexity of cancer is still hard to understand. The most remarkable characteristics of the genomic data are severely high-dimensional features with a small number of samples, such as gene expression data. The traditional data mining method has a limited ability to process these asymmetry datasets. In order to select the key genes from the high-dimensional gene expression data for cancer clustering analysis, gene selection based on proportional hazard model is applied in this paper. The proportional hazard model is a statistical approach used for survival risk analysis. The significant genes are selected for clustering analysis based on gene expression dataset. We demonstrate the effectiveness of this method on breast cancer and lung cancer. The experiments show a better cancer clustering result of separating of samples into distinct subclasses.

The quantity of web information is growing exponentially with time, and the challenge of acquiring information efficiently by personal search engine is increasingly complex. Personal preference is not easily described but can be observed from the examples given. Supervised Clustering with Support Vector Machines have been introduced to learn personal preference, however, it is not specialized and cannot be applied to the domain knowledge. In this paper, we introduce the ontology and semantic similarity into SVM as similar measurement. Experiments on agricultural information acquisition validate that it is highly effective.

Conventional pests image classification methods may not be accurate due to the complex farmland background, sunlight and pest gestures. To raise the accuracy, the deep convolutional neural network (DCNN), a concept from Deep Learning, was used in this study to classify crop pests image. On the ground of our experiments, in which LeNet-5 and AlexNet were used to classify pests image, we have analyzed the effects of both convolution kernel and the number of layers on the network, and redesigned the structure of convolutional neural network for crop pests. Further more, 82 common pest types have been classified, with the accuracy reaching 91%. The comparison to conventional classification methods proves that our method is not only feasible but preeminent.

In this paper, the driving knowledge base of autonomous vehicle is designed. Based on the driving knowledge modeling system, the driving knowledge of autonomous vehicle is visually acquired, managed, stored, and maintenanced, which has vital significance for creating the development platform of intelligent decision-making systems of automatic driving expert systems for autonomous vehicle.

Ontology is widely used in semantic computing and reasoning, and various biomedicine ontologies have become institutionalized to make the heterogeneous knowledge computationally amenable. Relation words, especially verbs, play an important role when describing the interaction between biological entities in molecular function, biological process, and cellular component; however, comprehensive research and analysis are still lacking. In this article, we propose an automatic method to build interaction relation ontology by investigating relation verbs, analyzing the syntactic relation of PubMed abstracts to perform relation vocabulary expansion, and integrating WordNet into our method to construct the hierarchy of relation vocabulary. Five attributes are populated automatically for each word in interaction relation ontology. As a result, the interaction relation ontology is constructed; it contains a total of 963 words and covers the most relation words used in existing methods of proteins interacti...

This paper proposes an updated prediction ARMA (autoregressive moving average) model for the disadvantage of traditional model that the future value forecasted by k-step ahead predictive model from time t didn't include the newest information on time t + 1 with the passage of time after a model was build. For this purpose, we adapt an approach of combining the ARMA model's difference equation form and transfer form (with Green's function) to achieve that new prediction value will calculate the change of the newest observation instead of reestablishing a new model. Furthermore, this method obtains higher forecasting accuracy and less computation. Finally we take an experiment on a time series sequence data to indicate the model's efficiency and effectiveness.

Sharing nearest neighbor (SNN) is a novel metric measure of similarity, and it can conquer two hardships: the low similarities between samples and the different densities of classes. At present, there are two popular SNN similarity based clustering methods: JP clustering and SNN density based clustering. Their clustering results highly rely on the weighting value of the single edge, and thus they are very vulnerable. Motivated by the idea of smooth splicing in computing geometry, the authors design a novel SNN similarity based clustering algorithm within the structure of graph theory. Since it inherits complementary intensity-smoothness principle, its generalizing ability surpasses those of the previously mentioned two methods. The experiments on text datasets show its effectiveness.

Discovering causal relationships in large databases of observational data is challenging. The pioneering work in this area was rooted in the theory of Bayesian network (BN) learning, which however, is a NP-complete problem. Hence several constraint-based algorithms have been developed to efficiently discover causations in large databases. These methods usually use the idea of BN learning, directly or indirectly, and are focused on causal relationships with single cause variables. In this paper, we propose an approach to mine causal rules in large databases of binary variables. Our method expands the scope of causality discovery to causal relationships with multiple cause variables, and we utilise partial association tests to exclude noncausal associations, to ensure the high reliability of discovered causal rules. Furthermore an efficient algorithm is designed for the tests in large databases. We assess the method with a set of real-world diagnostic data. The results show that our method can effectively discover interesting causal rules in large databases. Refereed/Peer-reviewed

Cloud computing is a network-based computing model; it provides support for the upper cloud application environment by building a cloud computing infrastructure. The combination of GIS and cloud computing can provide new prospects for the development of information storage, processing and its application for the GIS. This paper presents a GIS application model based on cloud computing. The model which based on cloud computing, data storage and transparent user services uses GIS infrastructure and service facilities as the service support platform. It provides users with a stable, efficient and reliable service.

The large amount of entity data are continuously published on web pages. Extracting these entities automatically for further application is very significant. Rule-based entity extraction method yields promising result, however, it is labor-intensive and hard to be scalable. The paper proposes a web entity extraction method based on entity attribute classification, which can avoid manual annotation of samples. First, web pages are segmented into different blocks by algorithm Vision-based Page Segmentation (VIPS), and a binary classifier LibSVM is trained to retrieve the candidate blocks which contain the entity contents. Second, the candidate blocks are partitioned into candidate items, and the classifiers using LibSVM are performed for the attributes annotation of the items and then the annotation results are aggregated into an entity. Results show that the proposed method performs well to extract agricultural supply and demand entities from web pages.

Among most of the subspace learning methods, Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) are classic ones. PCA tries to maximize the total scatter across all classes. In this case, however, the data set, with a small between-class scatter and a large within-class scatter, can also have a large total scatter. It conflicts with Maximum Margin Criterion (MMC) which tries to maximize the between-class scatter and minimize the within-class scatter. To address the conflict problem, we proposed a dynamic subspace learning method which can balance the objectives of PCA and MMC simultaneously by searching for the best coefficient. Our experiments are implemented by classification on two tumor microarray datasets. Firstly a simple t-test was used for gene selection, then our novel method was applied to gene extraction, and finally we adopted KNN and SVM classifiers to evaluate the effectiveness of our method. Results show that the new feature extractors are effective and stable.

In this paper we propose a novel approach of learning preference relations using Support Vector Regression (SVR). It answers the problem of consistent ranking and improves the ability of generalization to ranking for the property of SVR method. Meanwhile, the Wilcoxon-Mann-Whitney (WMW) statistic is introduced to evaluate the result of the ranking algorithm. The experiments on an artificial dataset and some benchmark datasets show the effectiveness of the proposed algorithm. An application to ranking in web searching system based on the proposed method is also demonstrated.

Visualizing the result of clustering is an important part of pattern recognition research. A novel method of visualizing the result of fuzzy clustering base on GIS is proposed, which doesn’t involve dimensionality reduction so that the key features are not missed. The experiment on real dataset shows its effectiveness.

A novel method is presented in this paper to study the use of SVM classifiers for multiple feature classification. While commonly multiple binary SVM classifiers are trained on features individually and the outputs of the classifiers are linearly combined for multiple feature classification, our method trains and combines these classifiers simultaneously with lower complexity. To obtain the optimal/suboptimal weights of different classifiers, an efficient algorithm is developed to takes into account both a base classifier's performance on the training data and its generalization ability, while traditional combination approaches consider a base classifierpsilas performance only. Experiments were performed and the results demonstrate the effectiveness and efficiency of the novel approach.

In this paper, a novel algorithm of mining multidimensional association rules in relational database is proposed. First, design a particular structure including many indexes to save multidimensional itemsets, which makes time of finding itemsets decrease. Basing on statistic idea, the algorithm saves all frequent 1-itemsets and their support members at scanning the database for the first time. Frequent k-itemsets will be generated with no necessary to scan the database more times because of using these support members. Compared with some traditional algorithms of mining association rules, the algorithm presented in this paper has better executive efficiency and expansibility, which is proved in our experiments.

Fuzzy ellipsoid theory was introduced to mining fuzzy association rule from quantitative data. The algorithm includes two steps. First, fuzzy clustering procedure is taken to get fuzzy ellipsoids of the feature space. Second, mining procedure is taken to get fuzzy association rule based on the projection regions of fuzzy ellipsoids. The algorithm was compared with a Apriori-based fuzzy association rule mining algorithm. Results show that the algorithm is considerably quicker, especially when mining from large databases. Experiments on real-life data also show that the algorithm is effective to find association rules.

A genetic weighted fuzzy rule-based system is proposed in this paper, in which the parameters of membership functions including position and shape of the fuzzy rule set and weights of rules are evolved using a genetic algorithm. The efficiency of the system has been illustrated in the process of classifying the iris data. This paper also illustrates that compared with non-weighted fuzzy rules, weighted fuzzy rules can lead to better fuzzy system.

Dimension reduction methods including feature selection and feature extraction have played an important role in data mining and pattern recognition. In this study, we propose a novel unsupervised feature selection approach based on sparse representation theory, namely Sparsity Score (SS). Due to the sparse representation procedure, SS not only owns the global property of Variance Score (VS) and the local property of Laplacian Score (LS), but also possesses the discriminating nature. Experimental results, based on three well-known face datasets (Yale, ORL and CMU PIE), reveal that SS performs well in the evaluation of the feature significance, and it significantly outperforms VS and LS.

Mining Protein-Protein Interactions (PPIs) from the fast-growing biomedical literature resources has been proven as an effective approach for the identification of biological regulatory networks. This paper presents a novel method based on the idea of Interaction Relation Ontology (IRO), which specifies and organises words of various proteins interaction relationships. Our method is a two-stage PPI extraction method. At first, IRO is applied in a binary classifier to determine whether sentences contain a relation or not. Then, IRO is taken to guide PPI extraction by building sentence dependency parse tree. Comprehensive and quantitative evaluations and detailed analyses are used to demonstrate the significant performance of IRO on relation sentences classification and PPI extraction. Our PPI extraction method yielded a recall of around 80% and 90% and an F1 of around 54% and 66% on corpora of AIMed and BioInfer, respectively, which are superior to most existing extraction methods.