Search

Orchid is a kind of terrestrial herb and it has elegant flower posture, quiet flower fragrance, rich colors and noble moral, therefore it has high ornamental value and is deeply loved by people. There are many kinds of orchids, and some of them are similar in shape, texture and color, which make people difficult to quickly and correctly distinguish them. As the existing classification model of orchid species have the problems of low accuracy rate and long classification time because of the inter species similarities and intra species differences in orchid species, thus influencing its wide application. In order to solve the problem above, in this paper, an improved classification model based on feature fusion is proposed for orchid species. The achievement of the paper lies in the fact that we successfully developed a classification model based on feature fusion to realize the high-efficient classification for orchid species. Specifically, in our scheme, firstly we obtained 12 orchid image sets with number of 12,227 images by network and field photography; Secondly we analyzed and studied the semantic relationship of different scale features from acquired orchid images above; Thirdly we designed an improved classification model based on feature fusion on the basis of the semantic relationship above; At last, we used the classification model above to realize the high-efficient classification for 12 orchid species. The experimental results showed that our proposed classification model based on feature fusion in this paper can realize 92.98% classification accuracy rate compared with classification models without using feature fusion technology, which can greatly improve the classification efficiency for orchid species.

Zero-shot learning (ZSL) aims to recognize objects from unseen classes only based on labeled images from seen classes. Most existing ZSL methods focus on optimizing feature spaces or generating visual features of unseen classes, both in conventional ZSL and generalized zero-shot learning (GZSL). However, since the learned feature spaces are suboptimal, there exists many virtual connections where visual features and semantic attributes are not corresponding to each other. To reduce virtual connections, in this paper, we propose to discover comprehensive and fine-grained object parts by building explanatory graphs based on convolutional feature maps, then aggregate object parts to train a part-net to obtain prediction results. Since the aggregated object parts contain comprehensive visual features for activating semantic attributes, the virtual connections can be reduced by a large extent. Since part-net aims to extract local fine-grained visual features, some attributes related to global structures are ignored. To take advantage of both local and global visual features, we design a feature distiller to distill local features into a master-net which aims to extract global features. The experimental results on AWA2, CUB, FLO, and SUN dataset demonstrate that our proposed method obviously outperforms the state-of-the-arts in both conventional ZSL and GZSL tasks.

Honey manufacturers and traders have suffered economic damage due to the spurious labeling of honey botanical sources. The traditional detection methods for distinguishing honey origins have inherent performance disadvantages, such as costly equipment, complicated operations, and need more time. This study proposes a novel method for identifying the origin of honey by applying an electronic tongue (ET) and deep learning algorithm. First, a self-developed ET device is utilized to collect unique fingerprints of honey samples from different botanical origins. Due to the high dimensionality and complex nonlinearity of the collected signal, an unsupervised convolutional autoencoder (CAE) is developed to reduce the dimension of the ET signals and achieve effective feature extraction. Moreover, in order to avoid the gradient disappearance and suppress redundant information, the self-adaptive attention mechanism and residual learning are introduced in the CAE model to improve the feature extraction ability of the model. The proposed model is further used to measure actual honey samples and contrast with multiple machine learning and deep learning models. The experimental results show that the proposed model demonstrates better classification capabilities for honey from different botanical sources, and the accuracy, precision, F1, and area under the curve (AUC) score reach 99.25%, 99.14%, 0.9919, and 0.9957, respectively. Compared with the other detection methods, the proposed method exhibited advantages in terms of convenience and accuracy of detection, which has great application prospects in the field of on-site measurements.

The adulteration and counterfeiting of tea products lead to economic losses and poses health concerns for consumers. This study proposed a novel method to rapidly discriminate the category of oolong tea, a famous Chinese semifermented tea, by applying an electronic nose (E-nose) and electronic tongue (E-tongue) system combined with a modified generative adversarial network (GAN) - temporal convolutional network (TCN) composite model. Specifically, the olfactory and gustatory fingerprint signals of tea samples are first collected by E-nose and E-tongue sensory systems, respectively. Considering the unbalanced scarcity and small-sample nature of data collection, a Wasserstein GAN model is employed to learn the data representations of E-nose and E-tongue signals and generate highly realistic training samples. Two improved temporal convolutional networks (TCNs) integrated with a squeeze-and-excitation attention module are proposed to learn the significant features and discover the internal patterns from the signals of the E-nose and E-tongue. Then, a dynamic fusion module (DFM) is developed to fuse the features of the E-nose and E-tongue to achieve information complementarity and enhancement. Finally, the fused information is sent to a classifier to predict the class label. The proposed approach is assessed using different performance metrics, and the experimental results reveal that an accuracy of 99.33% was achieved. The above research will provide a new approach for the rapid identification of different categories of oolong tea, which has a wide range of application prospects in the rapid classification and detection of other agricultural products.

The determination of parameters is a key issue in discrete element simulation and controls the accuracy and reliability of the simulation results. In this paper, a sandy loam soil with three water contents is studied and soil particles based on the EEPA model are modelled in three shapes (spheres, columns and Triangular pyramid) to accurately represent the actual soil particle. The sensitivity of the input parameters in the EEPA model is investigated by the Plackett–Burman test. The results show that the coefficient of static friction, coefficient of rolling friction and surface energy between soil particles have a highly significant effect on the angle of repose, and the coefficient of restitution has a significant effect on the angle of repose. The sensitivity parameters are calibrated by the central combination test, and the optimal combination of parameters is obtained. The accuracy of the parameters calibrated is validated by comparing the simulation results of the direct shear test with the actual test results. Therefore, the parameters calibrated satisfy both the flow and mechanical properties of the particles.

Approaches based on near infrared hyperspectral imaging (NIR-HSI) technology combined with machine learning have been developed to classify the leaves of hybrid cherry tomatoes and then identify the species of hybrid cherry tomato plants. The near infrared (NIR) hyperspectral images of 400 cherry tomato leaves (100 per species) were collected in the wavelength range of 900–1700 nm. Machine learning algorithms such as linear discriminant analysis (LDA), random forest (RF), and support vector machine (SVM) were employed to construct leaf classification models with the hyperspectral data preprocessed by Savitzky-Golay (SG) smoothing filter, first derivative (first Der) and standard normal variate (SNV). Principle of Component Analysis (PCA) was also used to reduce the data dimension and extract spectral features. It is revealed that the LDA model reaches the highest classification accuracy among the three machine learning algorithms and SNV can lead to higher improvement in model accuracy than other preprocessing methods of SG smoothing and first Der. Analysis based on PCA spectral feature extraction demonstrates that differences occur in internal material content in the leaves of cherry tomato plants with different species, which renders the models being able to distinguish between the species. Another important work was performed to reveal the different effects of the mesophyll and vein regions (VR) on the accuracy of the leaf classification model. It is demonstrated that the classification accuracy is improved by a value of 0.033 or 0.042 when mesophyll substitutes vein or whole leaf as regions of interest (ROI) to extract reflectance spectra for modeling. As a result, the accuracy of the training and test set respectively reached a high value of 0.998 and 0.973 for the LDA classification model combined with the SNV preprocessing method. The results propose that the use of mesophyll region (MR) as ROI can improve the performance of the leaf classification model, which provides a new strategy for efficient and non-destructive classification of different hybrid cherry tomato plants.

ABSTRACTPine wilt disease (PWD) is one of the most destructive forest diseases in the world. Therefore, timely monitoring of PWD is essential to the preservation of the ecological environment. However, the complex topography of PWD-outbreak locations affords many limitations in the manual detection of the disease. The use of unmanned aerial vehicles (UAVs) and deep learning technology to detect PWD-infected trees has gained popularity in recent years. In this study, we configured the You Only Look Once version 3 (YOLOv3) model according to the characteristics of the three disease stages of PWD and proposed an improved model called Effi_YOLO_v3. The results revealed that the improved model achieved good detection performance. The proposed model yielded a mean average precision (mAP) of 94.39%, and the classification of the different infection stages was relatively accurate. The recall values for the classifications of trees in the early-infection, late-infection, and death stages were 89.15%, 86.13%, and 86.77%, respectively. This indicates that the model offers good applicability in detecting different stages of PWD in trees.

Deep neural networks (DNNs) are easily fooled by adversarial examples. Most existing defense strategies defend against adversarial examples based on full information of whole images. In reality, one possible reason as to why humans are not sensitive to adversarial perturbations is that the human visual mechanism often concentrates on most important regions of images. A deep attention mechanism has been applied in many computer fields and has achieved great success. Attention modules are composed of an attention branch and a trunk branch. The encoder/decoder architecture in the attention branch has potential of compressing adversarial perturbations. In this article, we theoretically prove that attention modules can compress adversarial perturbations by destroying potential linear characteristics of DNNs. Considering the distribution characteristics of adversarial perturbations in different frequency bands, we design and compare three types of attention modules based on frequency decomposition and reorganization to defend against adversarial examples. Moreover, we find that our designed attention modules can obtain high classification accuracies on clean images by locating attention regions more accurately. Experimental results on the CIFAR and ImageNet dataset demonstrate that frequency reorganization in attention modules can not only achieve good robustness to adversarial perturbations, but also obtain comparable, even higher classification, accuracies on clean images. Moreover, our proposed attention modules can be integrated with existing defense strategies as components to further improve adversarial robustness.

We propose a hybrid algorithm based on the nested genetic algorithm (GA) and integer particle swarm optimization (PSO) for multiple unmanned aerial systems (multi-UASs) plant protection operations in multiple segmented fields, with multibases loading pesticides and power. In the proposed algorithm, a preprocessing phase is adopted that includes segmented fields splitting and merging, to shrink the operation field number (i.e., multi-UASs total sorties) and ensure that each generated field area remains smaller than (but close to) the unmanned aerial system (UAS) maximum single-sortie operation area. An improved one-sortie coverage path generation model is established, avoiding power waste in the case when UAS needs to operate on more than one subfield in one sortie. The integer PSO is then used to optimize the initial assignment for UASs on multibases, with the purpose of reducing total operation time and nonoperation flight distance of overall coverage path planning. The proposed algorithm solves the problems of how to design pesticide spraying coverage path on segmented agricultural fields with several bases, and allocate to multiple plant protection UASs with constrained loading capacity. Computer simulations are provided to validate the effectiveness of our algorithms and the performance compared with other algorithms. The test results show that the total number of flight sorties using nested GA (53) is smaller than that using some representative metaheuristic algorithms (57), density-based spatial clustering of applications with noise (60), and traditional planning methods (83). Both nonoperation distance and total operation time obtained by nest GA and integer PSO are smaller than those using other planning methods. The appearance of multibases loading pesticide and power, would reduce the nonspraying flight energy and operation time wastage of UASs.

To improve the efficiency of innovative design and performances of a product, a self-commentary design method facing product-oriented innovation and fuzzy back end evaluation was proposed. Firstly, a functional structure mode of product was built on the basis of axiomatic design theory. The contradiction matrix must be built and the initial scheme should be redesigned by TRIZ conflict analysis method in order to be decoupled, and got product innovation program. Lastly, taking the lower extremity exoskeleton robot as an example, the structure was decoupled and optimised using Altium Designer (AD) and TRIZ theory. The performance of the lower extremities exoskeleton robot was evaluated quantificationally by using fuzzy analytic hierarchy process, which verified the feasibility and effectiveness of the decoupling design and fuzzy back end evaluation method for product innovation design and inspection.

The unique α-Fe2O3hexagonal nanosheets have been successfully prepared through a facile solvothermal method with subsequent calcining process. The as-synthesized α-Fe2O3sample exhibits p-type conductive performance. The gas sensitivities of the as-fabricated Fe2O3nanosheets-based sensor are comprehensively studied and show good gas-sensing performance toward aniline. Furthermore, the aniline-sensing mechanism is investigated and the origin of the p-type conductivities is also discussed.

Quantitative inversion is a significant topic in remote sensing science. The development of visible light-based hyperspectral reconstruction techniques has opened up novel prospects for low-cost, high-precision remote sensing inversion in agriculture. The aim of this study was to assesses the effectiveness of hyperspectral reconstruction technology in agricultural remote sensing applications. Hyperspectral images were reconstructed using the MST++ hyperspectral reconstruction model and compared with the original visible light images in terms of their correlation with physiological parameters, the accuracy of single-feature modeling, and the accuracy of combined feature modeling. The results showed that compared to the visible light image, the reconstructed data exhibited a stronger correlation with physiological parameters, and the accuracy was improved in both the single-feature and the combined feature inversion mode. However, compared to multispectral sensors, hyperspectral reconstruction provided limited improvement on the inversion model accuracy. It was concluded that for physiological parameters that are not easy to be directly observed, deep mining of features in visible light. data through hyperspectral reconstruction technology can improve the accuracy of the inversion model. Appropriate feature selection and simple models are more suitable for the remote sensing inversion task of traditional agronomic plot experiments. To strengthen the application of hyperspectral reconstruction technology in agricultural remote sensing, further development is necessary with broader wavelength ranges and more diverse agricultural scenes.

ABSTRACTSpider mites are important pests that cause severe economic damage to cotton. They feed on underside of leaves, piercing the chloroplast-containing cells, resulting in foliar damage and yield reduction. This paper proposed a two-stage classification approach for mite-infestation detection based on machine learning methods. Two cotton fields were selected for study, and the UAV imagery collection and concurrent ground investigation were conducted on July 20-21th, 2017. Mosaicking and geo-registration were performed on the collected multispectral imagery. Support Vector Machine (SVM) was used for scene classification, and a transferred Convolutional Neural Network (CNN) was applied for mite-infestation identification. Experimental results showed that our method outperformed others in terms of accuracy, which demonstrated that our approach has potential in mite-infestation detection using UAV multispectral imagery.

This article describes how quickly picking up some valuable information from massive RFID event stream often faces with the problem of long detection time, high memory consumption and low detection efficiency due to its stream characteristics of volume, velocity, variety, value and veracity. Aim to solving the problems above, an efficient complex event processing method based on NFA-HTBTS (Nondeterministic Finite Automaton-Hash Table B+ Tree Structure) is presented in this article. The achievement of this article lies in that we successfully use the union of NFA and HTBTS to realize the detection of complex event in massive RFID event stream. Specially, in our scheme, after using NFA to match related primitive events from massive RFID event stream, we use hash table and B+ tree structure to successfully realize the detection of complex event from large matched results above, as a result, these problems existed in current methods above can be effectively solved by our scheme. The simulation results show that our proposed scheme outperforms some general methods for massive RFID event stream.

Soluble solids content (SSC) is one of most important quality indicators of the navel orange. In order to explore the feasibility of SSC prediction for the navel orange from different origins using near infrared (NIR) spectroscopy, we collected seven groups Newhall navel orange (Citrus sinensis(L) Osb.) samples from seven origins (i.e. Beibei, Fengjie, Leibo, Linhai, Wusheng, Xinfeng, and Yizhang) in China, and all the samples were combined as the eighth group. The difference of the growing environments caused the variation of SSC of oranges from different origins. The partial least squares regression (PLS) models were applied to predict the SSC of its origin samples and other origin samples. The results predicted by origin-model were the best compared to cross-origin-prediction results. So it was necessary to recognize origin as the first steps of SSC prediction. Linear discriminant analysis (LDA) model with the top 18 principle components could recognize the origins of samples with 100% accuracy. The overall results demonstrated that it was feasible that SSC prediction for navel orange based on origin recognition using NIR spectroscopy.

In recent years, the fake Xihu Longjing Tea has damaged its brand image and reputation. This paper based on the Fisher's discriminant analysis using the fixed-size moving window evolving factor analysis to find characteristic spectra through analyzing the near infrared spectroscopy of Xihu Longjing Tea and Zhejiang Longjing Tea. The Fisher's discriminant analysis was used to reduce the data dimension combined with the principal component analysis. A discrimination model was set up for the identification of the Xihu Longjing Tea and Zhejiang Longjing Tea. The model's accuracy is 97.3%. The results proved that this model is feasible to identify the differences between the Xihu Longjing Tea and Zhejiang Longjing Tea. Unlike other methods, the tea does not need to be made into a powder. It also lays out a theoretical foundation for developing an identification instrument for Xihu Longjing Tea.

Samples of pheromone lures used in boll weevil, Anthonomus grandis(Boheman), eradication programs are routinely analyzed by Gas Chromatography (GC) to ensure lures are adequately dosed with grandlure, the synthetic aggregation pheromone produced by male weevils. However, preparation of GC samples is tedious, time consuming, and requires a moderate level of experience. We examined the use of a commercially-available electronic nose (e-nose) for rapidly assessing the grandlure contents of lures. The e-nose was trained to recognize headspace collections of grandlure emitted from new lures and after lures were aged under field conditions for 4 d, 7 d, 10 d, and 14 d. Based on cross-validation of the training set, the e-nose was 82% accurate in discriminating among the different age classes of lures. Upon sampling headspace collections of pheromone from a different set of field-aged lures, the e-nose was <50% accurate in discriminating 4 d, 7 d, and 10 d aged lures from the other age classes of lures. However, the e-nose identified new and 14 d aged lure samples with 100% accuracy. In light of these findings, e-nose technology shows considerable promise as an alternative approach for rapidly assessing the initial grandlure contents of lures used in boll weevil eradication programs.

This paper presents a new Pulse Width Modulation (PWM) controller for Unmanned Aerial Vehicle (UAV) precision sprayer for agriculture using a TL494 fixed-frequency pulse width modulator together with a data acquisition board and developed software. An UAV can be remotely controlled or flown autonomously by pre-programmed flight plans. The PWM controller was implemented through the guidance system on the UAV with control commands sent between the UAV helicopter and the ground control station via a wireless telemetry system. The PWM controller was tested and validated using LabVIEW 8.2. Several analyses were performed in a laboratory to test different control signals. The results show that the PWM controller has promise as a higher precision technique for spray applications, which will improve efficiency of pesticide application, especially in crop production areas.

Thermal conductivities of four types of granular rice starches, regular, low protein, waxy and low protein waxy, were measured at bulk densities of 600 and 800 kg/m3, temperatures of 20, 50 and 80 °C, and moisture contents of 4, 20 and 40% (wet basis), using a line-source thermal probe system equipped with a computerized data acquisition system. A completely randomized experimental design with factorial treatment structure resulted in a total of 72 treatments, each with 3 replications. The thermal conductivity was significantly different among various rice starches. Low protein and waxy rice starches had higher thermal conductivities than regular and low protein waxy rice starches. Thermal conductivity increased significantly with increases in bulk density, temperature and moisture content. Prediction models were developed to calculate thermal conductivity values for practical applications.

Thermal properties of native tapioca starch, including thermal conductivity, specific heat, and thermal diffusivity at 4, 15, and 30% moisture contents, and 25, 50, and 75°C temperature levels were studied. Thermal conductivity was measured using a line-heat source thermal probe equipped with a computerized data acquisition system; specific heat was measured using Differential Scanning Calorimetry (DSC) and thermal diffusivity was calculated from the thermal conductivity, density, and specific heat. Thermal conductivity, specific heat, and thermal diffusivity increased with the increase of temperature and/or moisture content. Prediction models for the thermal properties of tapioca starch were developed for practical applications.

Starch and starch/fructose samples were prepared using ct‐amylase to hydrolyze starch and formed in a mold to produce 60 × 20 × 5 mm rectangular parallel‐piped specimens. Drying experiments using three levels of initial sample moisture content (48, 54, 60% wb). and two levels of drying temperature (48, 88C) were conducted using a computer‐controlled connective oven. Results indicated that the