High performance vegetable classification from images based on AlexNet deep learning model

Deep learning techniques can automatically learn features from a large number of image data set.Automatic vegetable image classification is the base of many applications.This paper proposed a high performance method for vegetable images classification based on deep learning framework.The AlexNet network model in Caffe was used to train the vegetable image data set.The vegetable image data set was obtained from ImageNet and divided into training data set and test data set.The output function of the AlexNet network adopted the Rectified Linear Units(ReLU)instead of the traditional sigmoid function and the tanh function,which can speed up the training of the deep learning network.The dropout technology was used to improve the generalization of the model.The image data extension method was used to reduce overfitting in the learning process.With AlexNet network model used for training different number of vegetable image data set,the experimental results showed that the classification accuracy decreases as the number of data set decreases.The experimental verification indicated that the accuracy rate of the deep learning method in the test data set reached as high as 92.1%,which was greatly improved compared with BP neural network(78%)and SVM classifier(80.5%)methods.

A Scheme of Vegetation Classification of Taiwan, China

The complexity of natural conditions leads to the complexity of vegetation types of Taiwan of China, which has both tropical and cold-temperate vegetation types, and could be depicted as the vegetation miniature of China or even for the world. The physiognomic-floristic principle was adopted for the vegetation classification of Taiwan. The units of rank from top to bottom are: class of vegetation-type, order of vegetation-type, vegetation-type, alliance group, alliance and association. The high-rank units (class, order and vegetation-type) are classified by ecological physiognomy, while the median and lower units by the species composition of community. At the same time the role of dominant species and character species will also be considered. The dominant species are the major factor concerned with the median ranks (alliance group, and alliance) because they are the chief components of community, additionally their remarkable appearance is easy to identify; the character species (or diagnostic species) are for relatively low ranks (association) because they will clearly show the interspecies relation-ship and the characteristics of community. According to this principle, vegetation of Taiwan is classi-fied into five classes of vegetation-types (forests, thickets, herbaceous vegetation, rock fields vegetation, swamps and aquatic vegetation), 29 orders of vegetation-types (cold-temperate needle-leaved forests, cool-temperate needle-leaved forests, warm-temperate needle-leaved forests, warm needle-leaved forests, deciduous broad-leaved forests, mixed evergreen and deciduous broad-leaved forests, evergreen mossy forests, evergreen sclerophyllous forests, evergreen broad-leaved forests, tropical rain forests, tropical monsoon forests, coastal forests, warm bamboo forests, evergreen needle-leaved thickets, sclerophyllous thickets, deciduous broad-leaved thickets, evergreen broad-leaved thickets, xerothermic thorn-succulent thickets, bamboo thickets, meadows, sparse shrub grasslands, savannahic grasslands, sparse scree communities, chasmophytic vegetation, woody swamps, herbaceous swamps, moss bogs, fresh water aquatic vegetation, salt water aquatic vegetation) and 53 vegetation-types. The main alliances of each vegetation-type are described.

Fruits and Vegetables Freshness Categorization Using Deep Learning

The nutritional value of perishable food items,such as fruits and vegetables,depends on their freshness levels.The existing approaches solve a binary class problem by classifying a known fruit\vegetable class into fresh or rotten only.We propose an automated fruits and vegetables categorization approach that first recognizes the class of object in an image and then categorizes that fruit or vegetable into one of the three categories:purefresh,medium-fresh,and rotten.We gathered a dataset comprising of 60K images of 11 fruits and vegetables,each is further divided into three categories of freshness,using hand-held cameras.The recognition and categorization of fruits and vegetables are performed through two deep learning models:Visual Geometry Group(VGG-16)and You Only Look Once(YOLO),and their results are compared.VGG-16 classifies fruits and vegetables and categorizes their freshness,while YOLO also localizes them within the image.Furthermore,we have developed an android based application that takes the image of the fruit or vegetable as input and returns its class label and its freshness degree.A comprehensive experimental evaluation of proposed approach demonstrates that the proposed approach can achieve a high accuracy and F1score on gathered FruitVeg Freshness dataset.The dataset is publicly available for further evaluation by the research community.

Classification of Vegetation in North Tibet Plateau Based on MODIS Time-Series Data

Based on the 16d-composite MODIS （moderate resolution imaging spectroradiometer）-NDVI（normalized difference vegetation index） time-series data in 2004, vegetation in North Tibet Plateau was classified and seasonal variations on the pixels selected from different vegetation type were analyzed. The Savitzky-Golay filtering algorithm was applied to perform a filtration processing for MODIS-NDVI time-series data. The processed time-series curves can reflect a real variation trend of vegetation growth. The NDVI time-series curves of coniferous forest, high-cold meadow, high-cold meadow steppe and high-cold steppe all appear a mono-peak model during vegetation growth with the maximum peak occurring in August. A decision-tree classification model was established according to either NDVI time-series data or land surface temperature data. And then, both classifying and processing for vegetations were carried out through the model based on NDVI time-series curves. An accuracy test illustrates that classification results are of high accuracy and credibility and the model is conducive for studying a climate variation and estimating a vegetation production at regional even global scale.

Using classification assignment rules to assess land-use change impacts on forest biodiversity at local-to-national scales

Background： Ecosystem representation is one key component in assessing the biodiversity impacts of land-use changes that will irrevocably alter natural ecosystems. We show how detailed vegetation plot data can be used to assess the potential impact of inundation by a proposed hydroelectricity dam in the Mokihinui gorge, New Zealand, on representation of natural forests. Specifically we ask： 1） How well are the types of forest represented Locally, regionally, and nationally; and 2） How does the number of distinct communities （i.e. beta diversity） in the target catchment compare with other catchments nationally？ Methods： For local and regional comparisons plant species composition was recorded on 45 objectively located 400 m2 vegetation plots established in each of three gorges, with one being the proposed inundation area of the Mokihinui lower gorge. The fuzzy classification framework of noise clustering was used to assign these plots to a specific alliance and association of a pre-existing national-scale classification. NationaLly, we examined the relationship between the number of alliances and associations in a catchment and either catchment size or the number of plots per catchment by fitting Generalised Additive Models. Results： The four alliances and five associations that were observed in the Mokihinui lower gorge arepresent in the region but limited locally. One association was narrowly distributed nationally, but is the mostfrequent association in the Mokihinui lower gorge; inundation may have consequences of national importance to its long-term persistence. That the Mokihinui lower gorge area had nearly twice as many plots that could not be assigned to pre- existing alliances and associations than either the Mokihinui upper or the Karamea lower gorges and proportionally more than the national dataset emphasises the compositional distinctiveness of this gorge. These outlier plots in the Mokihinui lower gorge may be unsorted assemblages of species or reflect sampling bias or that native- dominated woody riparian vegetation is rare on the landscape. At a national scale, the Mokihinui catchment has a higher diversity of forest alliances and associations （i.e. beta-diversity） than predicted based on catchment size and sampling intensity. Conclusions： Our analytical approach demonstrates one transparent solution to a common conservation planning problem： assessing how well ecosystems that will be destroyed by a proposed land-use change are represented using a multi-scale spatial and compositional framework. We provide a useful tool for assessing potential consequences of land-use change that can help guide decision making.

Classification of vegetative types in Changbai Mountain based on optical and microwave remote sensing data

Highly accurate vegetative type distribution information is of great significance for forestry resource monitoring and management.In order to improve the classification accuracy of forest types,Sentinel-1 and 2 data of Changbai Mountain protection development zone were selected,and combined with DEM to construct a multi-featured random forest type classification model incorporating fusing intensity,texture,spectral,vegetation index and topography information and using random forest Gini index(GI)for optimization.The overall accuracy of classification was 94.60%and the Kappa coefficient was 0.933.Comparing the classification results before and after feature optimization,it shows that feature optimization has a greater impact on the classification accuracy.Comparing the classification results of random forest,maximum likelihood method and CART decision tree under the same conditions,it shows that the random forest has a higher performance and can be applied to forestry research work such as forest resource survey and monitoring.

Modelling analysis embodies drastic transition among global potential natural vegetations in face of changing climate

Potential natural vegetation(PNV)is a valuable reference for ecosystem renovation and has garnered increasing attention worldwide.However,there is limited knowledge on the spatio-temporal distributions,transitional processes,and underlying mechanisms of global natural vegetation,particularly in the case of ongoing climate warming.In this study,we visualize the spatio-temporal pattern and inter-transition procedure of global PNV,analyse the shifting distances and directions of global PNV under the influence of climatic disturbance,and explore the mechanisms of global PNV in response to temperature and precipitation fluctuations.To achieve this,we utilize meteorological data,mainly temperature and precipitation,from six phases:the Last Inter-Glacial(LIG),the Last Glacial Maximum(LGM),the Mid Holocene(MH),the Present Day(PD),2030(20212040)and 2090(2081–2100),and employ a widely-accepted comprehensive and sequential classification sy–stem(CSCS)for global PNV classification.We find that the spatial patterns of five PNV groups(forest,shrubland,savanna,grassland and tundra)generally align with their respective ecotopes,although their distributions have shifted due to fluctuating temperature and precipitation.Notably,we observe an unexpected transition between tundra and savanna despite their geographical distance.The shifts in distance and direction of five PNV groups are mainly driven by temperature and precipitation,although there is heterogeneity among these shifts for each group.Indeed,the heterogeneity observed among different global PNV groups suggests that they may possess varying capacities to adjust to and withstand the impacts of changing climate.The spatio-temporal distributions,mutual transitions and shift tendencies of global PNV and its underlying mechanism in face of changing climate,as revealed in this study,can significantly contribute to the development of strategies for mitigating warming and promoting re-vegetation in degraded regions worldwide.

Vegetation Mapping for Regional Ecological Research and Management: A Case of the Loess Plateau in China

Vegetation maps are fundamental for regional-scale ecological research. However, information is often not sufficiently up to date for such research. The Loess Plateau is a key area for vegetation restoration projects and a suitable area for regional ecological research. To carry out regional vegetation mapping based on the principles of hierarchical classification, object-oriented methods, visual interpretation, and accuracy assessment, this study integrated land cover, high-resolution remote sensing images, background environmental data, bioclimate zoning data, and field survey data from the Loess Plateau. To further clarify the implications of vegetation mapping, we compared the deviation of the 2015 vegetation map of the Loess Plateau(VMLP) and the widely used vegetation map of China(VMC)(1 : 1 000 000) for the expressed vegetation information and the evaluation of ecosystem services. The results indicated that 1) the vegetation of the Loess Plateau could be divided into 9 vegetation type groups and 18 vegetation types with classification accuracies of 87.76% and 83.97%, respectively;2) the distribution of vegetation had obvious zonal regularity;3) a deviation of 29.56 × 10^4 km^2 occurred when the vegetation coverage area was quantified with the VMC;4) the vegetation classification accuracy affected the ecosystem service assessment, the total water yield of the Loess Plateau calculated by the VMC and other required parameters was overestimated by 2.2 × 10^6 mm in 2015. Because vegetation mapping is a basic and important activity, that requires greater attention, this study provides supporting data for subsequent multivariate vegetation mapping and vegetation management for conservation and restoration.

Assessing vegetation community distribution characteristics and succession stages in mountainous areas hosting coming Winter Olympics Games

Alpine mountain ecosystem shows strong interactions between abiotic and biotic parameters.They also receive high attention from human activities(natural tourism,trekking,skiing,etc.).However,as the potential disturbance risk areas in alpine mountains are increasing,it is necessary to understand the relationship between environmental factors and plant communities.This is also the key consideration for the coming international events such as the Winter Olympic Games,which could generate uncontrolled ecosystem issues not previously studied.The Yin Mountains in Chongli district,Zhangjiakou City,Hebei Province,China will be the core area of the 2022 Winter Olympic Games.We hypothesize that disturbances will be caused,therefore,the previous relationships between the habitat factors and plant community and the main environmental limiting factors before hosting them must be assessed to design future restoration plans.Therefore,we used the two-way indicator species analysis(TWINSPAN)and market basket analysis(MBA)for vegetation classification in 91 sampling plots.Plant community and relationships among environmental variables(altitude,slope position,aspect,direction,inclination,soil porosity,soil bulk density,organic matter content and soil pH)were investigated through the trend correspondence(DCA)and canonical correspondence analyses(CCA).Also,the TWINSPAN was used to classify the vegetation into 6 different groups.CCA analysis showed that i)the spatial variation of soil moisture and the content of soil organic matter are the main factors limiting the development of shrub and herb communities;ii)the distribution of different forest communities was mainly affected by terrain factors(altitude,aspect and slope position);iii)the dynamic changes of vegetation communities in different altitudes were affected by the fluctuation of environmental factors and human disturbance,and the shrubs and herbaceous plants in mid-to-high altitude areas(above 1400 m)generally show the process of transformation from the pioneer community to transitional community in the competition.We concluded that under the strong interference of human activities in the core construction area of the Olympic venues,higher damage intensity and lower resilience in the low altitude area is observed compared with the pioneer community.Whereas in the low altitude area(below 1400m)with a fragile ecological environment,although the plant diversity and coverage are poor,the potential impact and damage degree of the Olympic Games are greatly reduced due to the distance from the construction area of the core venues and good resilience.This information can help land managers and policymakers to anticipate human disturbances on plant communities and support guiding the most efficient ecological restoration after the Winter Olympic Games in 2022.

Texture-based latent space disentanglement for enhancement of a training dataset for ANN-based classification of fruit and vegetables

The capability of Convolutional Neural Networks(CNNs)for sparse representation has significant application to complex tasks like Representation Learning(RL).However,labelled datasets of sufficient size for learning this representation are not easily obtainable.The unsupervised learning capability of Variational Autoencoders(VAEs)and Generative Adversarial Networks(GANs)provide a promising solution to this issue through their capacity to learn representations for novel data samples and classification tasks.In this research,a texture-based latent space disentanglement technique is proposed to enhance learning of representations for novel data samples.A comparison is performed among different VAEs and GANs with the proposed approach for synthesis of new data samples.Two different VAE architectures are considered,a single layer dense VAE and a convolution based VAE,to compare the effectiveness of different architectures for learning of the representations.The GANs are selected based on the distance metric for disjoint distribution divergence estimation of complex representation learning tasks.The proposed texture-based disentanglement has been shown to provide a significant improvement for disentangling the process of representation learning by conditioning the random noise and synthesising texture rich images of fruit and vegetables。

Application of aerial remote sensing in the study on vegetation in Guangzhou,China

ＡｐｐｌｉｃａｔｉｏｎｏｆａｅｒｉａｌｒｅｍｏｔｅｓｅｎｓｉｎｇｉｎｔｈｅｓｔｕｄｙｏｎｖｅｇｅｔａｔｉｏｎｉｎＧｕａｎｇｚｈｏｕ，ＣｈｉｎａＣｈｅｎＳｈｉｊｉｅ；ＨｅＳｈｉｇａｎ；ＹａｎｇＪｉｅｌｉｎ；ＷａｎｇＬｉａｎｇｐｉｎｇ（Ｄｅｐａｒｔｍｅｎ...

Vegetation classification at the association level under the China Vegetation Classification System: an example of six Stipa steppe formations in China

Aims The latest China Vegetation Classification System(China-VCS)for natural/semi-natural vegetation has eight hierarchical levels:Association<Association-group<Subformation<Formation<Formation-group<Vegetation-subtype<Vegetation-type<Vegetation-type-group.The classification is based on dominant species and their growth forms and has been completed at the formation level.The principal challenge today in Chinese vegetation classification is to develop the China-VCS at levels below the formation in a way that is consistent with current international standards.We explored the following question:how can existing vegetation plot data help develop the China-VCS and improve its compatibility with other international classification systems?Methods We compiled 401 plots having plant cover and/or aboveground biomass measurements collected in six Stipa steppe formations and divided them into those with cover data(299 plots)and/or biomass data(283 plots).We applied a combination of hierarchical clustering and ordination to partition the cover and biomass data sets into formations and constituent associations.We then used supervised noise clustering to improve the classification and to identify the core plots representing each association.Diagnostic species were also identified at both association and formation levels.Finally,we compared the classification results based on cover and biomass data sets and combined these results into a comprehensive classification framework for the six formations.Important Findings Our results using cover data were comparable with those using biomass data at both formation and association levels.Three Stipa formations were classified into associations based on cover data,two based on biomass data and one based on both biomass and cover data.Twenty-seven associations were defined and proposed within the six formations,using cover or biomass data as consistent classification sections(CCSs).Both dominant species in the dominant stratum and diagnostic species from multiple strata of the core plots were used to characterize vegetation types at both formation and association levels,improving the compatibility of our classification with the International Vegetation Classification.Temperature and precipitation were found to be important climatic factors determining the distribution pattern and species composition of Stipa-dominated vegetation.We propose a framework for plotbased vegetation classification in the China-VCS,using our work with Stipa-dominated steppe vegetation as an example.We applied the concept of CCS to make optimal use of available data representing both plant cover and biomass.This study offers a model for developing the China-VCS to the association level in a way that is consistent with current international standards.

Vegetation classification of East China with multi-temporal NOAA-AVHRR data

East China lies in the subtropical monsoon cli-matic zone and is dominated by subtropical evergreen broad-leaved forests,a unique vegetation type mainly dis-tributed in East Asia with the largest distribution in China.It is important to be able to monitor and estimate forest biomass and production,regional carbon storage,and global climate change impacts on these important vegetation types.In this paper,we used coarse resolution remote sensing data to identify the vegetation types in East China and developed a map of the spatial distribution of vegetation types in this region.Nineteen maximum normalized difference vegeta-tion index(NDVI)composite images(acquisition time span of 7 months from February to August),which were derived from 10 days National Oceanographic and Atmospheric Administration(NOAA)Advanced Very High Resolution Radiometer(AVHRR)channel 1 and channel 2 observa-tions,an unsupervised classification method,and the ISODATA algorithm were employed to identify the vegeta-tion types.To reduce the dimensions of the dataset resulted in a total of 28 spectral clusters of land-cover of which two clusters were urban/bare soil and water,the images were processed using principal component analysis(PCA).The 26 remaining spectral clusters were merged into six vegeta-tion types using the Chinese vegetation taxonomy system:evergreen broad-leaved forest,coniferous forest,bamboo forest,shrub-grass,aquatic vegetation,and agricultural vegetation.The spatial distribution and areal extent for the coniferous forests,shrub-grass,evergreen broad-leaved for-ests,and agricultural vegetation were calculated and com-pared with the Vegetation Atlas of China at a 1:1,000,000 scale.The spatial accuracy and the area accuracy for conif-erous forests,shrub-grass,evergreen broad-leaved forests,and agricultural vegetation were 79.2%,91.3%,68.2%and 95.9%and 92.1%,95.9%,63.8%and 90.5%,respectively.The spatial accuracy and area accuracy of the bamboo forest were 28.7%and 96.5%,respectively;the spatial accuracy of aquatic vegetation was 69.6%,but there was a significant difference in its area accuracy because image acquisition did not cover the full year.Our study demonstrated the fea-sibility of using NOAA-AVHRR to identify the different vegetation types in the subtropical evergreen broad-leaved forest zone in East China.The spatial location of the six identified vegetation types agreed with the actual geo-graphical distribution of the vegetation types in East China.

Modern pollen assemblages of the surface lake sediments from the steppe and desert zones of the Tibetan Plateau

Modern pollen analysis is the basis for revealing the palaeovegetation and palaeoclimate changes from fossil pollen spectra.Many studies pertaining to the modern pollen assemblages on the Tibetan Plateau have been conducted,but little attention has been paid to pollen assemblages of surface lake sediments.In this study,modern pollen assemblages of surface lake sediments from 34 lakes in the steppe and desert zones of the Tibetan Plateau are investigated and results indicate that the two vegetation zones are dominated by non-arboreal pollen taxa and show distinctive characteristics.The pollen assemblages from the desert zone contain substantially high relative abundance of Chenopodiaceae while those from the steppe zone are dominated by Cyperaceae.Pollen ratios show great potential in terms of separating different vegetation zones and to indicate climate changes on the Tibetan Plateau.The Artemisia/Chenopodiaceae ratio and arboreal/non-arboreal pollen ratio could be used as proxies for winter precipitation.Artemisia/Cyperaceae ratio and the sum of relative abundance of xerophilous elements increase with enhanced warming and aridity.When considering the vegetation coverage around the lakes,hierarchical cluster analysis suggests that the studied sites can be divided into four clusters:meadow,steppe,desert-steppe,and desert.The pollen-based vegetation classification models are established using a random forest algorithm.The random forest model can effectively separate the modern pollen assemblages of the steppe zone from those of the desert zone on the Tibetan Plateau.The model for distinguishing the four vegetation clusters shows a weaker but still valid classifying power.It is expected that the random forest model can provide a powerful tool to reconstruct the palaeovegetation succession on the Tibetan Plateau when more pollen data from surface lake sediments are included.

Vegetation distribution pattern in the dam areas along middle-low reach of Lancang-Mekong River in Yunnan Province, China

Lancang-Mekong River Basin is one of ecoregions with rich biodiversity and high ecological values in the world. The basin has been strongly affected by human activities, particularly by dam construction. This study was conducted to investigate the vegetation distribution patterns in the dam areas along middle-low reach of the Lancang-Mekong River in Yunnan Province of China, where eight cascade dams have been planned or are being constructed. To identify the vegetation composition and structure, we sampled 126 quadrats along the transects arrayed vertically to both side of river channel from the year of 2004 to 2010. We found that the forest, shrub and grass communities were widely spread along the riverside. In low reach watershed of the Lancang-Mekong River, the dominated vegetations were grasses and shrubs which were severely disturbed by human activity. In middle reach of the Lancang-Mekong River, the dry-hot valley vegetation was found in the low valley. At high altitude, the pine forest and semi-evergreen seasonal forest were found. As a result of dam construction and operation, the structure and compositions of riparian vegetation were strongly changed. Some plants declined or disappeared due to the alteration of their habitats. The protection or restoration interventions are urgently needed to mitigate the risk of vegetation damage associated with dam projects along middle and low reach of the Lancang-Mekong River.

Combination of super-resolution reconstruction and SGA-Net for marsh vegetation mapping using multi-resolution multispectral and hyperspectral images

Vegetation is crucial for wetland ecosystems.Human activities and climate changes are increasingly threatening wetland ecosystems.Combining satellite images and deep learning for classifying marsh vegetation communities has faced great challenges because of its coarse spatial resolution and limited spectral bands.This study aimed to propose a method to classify marsh vegetation using multi-resolution multispectral and hyperspectral images,combining super-resolution techniques and a novel self-constructing graph attention neural network(SGA-Net)algorithm.The SGA-Net algorithm includes a decoding layer(SCE-Net)to preciselyfine marsh vegetation classification in Honghe National Nature Reserve,Northeast China.The results indicated that the hyperspectral reconstruction images based on the super-resolution convolutional neural network(SRCNN)obtained higher accuracy with a peak signal-to-noise ratio(PSNR)of 28.87 and structural similarity(SSIM)of 0.76 in spatial quality and root mean squared error(RMSE)of 0.11 and R^(2) of 0.63 in spectral quality.The improvement of classification accuracy(MIoU)by enhanced super-resolution generative adversarial network(ESRGAN)(6.19%)was greater than that of SRCNN(4.33%)and super-resolution generative adversarial network(SRGAN)(3.64%).In most classification schemes,the SGA-Net outperformed DeepLabV3+and SegFormer algorithms for marsh vegetation and achieved the highest F1-score(78.47%).This study demonstrated that collaborative use of super-resolution reconstruction and deep learning is an effective approach for marsh vegetation mapping.

Quantifying the characteristics of particulate matters captured by urban plants using an automatic approach

It is widely accepted that urban plant leaves can capture airborne particles. Previous studies on the particle capture capacity of plant leaves have mostly focused on particle mass and/or size distribution. Fewer studies, however, have examined the particle density, and the size and shape characteristics of particles, which may have important implications for evaluating the particle capture efficiency of plants, and identifying the particle sources. In addition, the role of different vegetation types is as yet unclear. Here, we chose three species of different vegetation types, and firstly applied an object-based classification approach to automatically identify the particles from scanning electron microscope（SEM）micrographs. We then quantified the particle capture efficiency, and the major sources of particles were identified. We found（1） Rosa xanthina Lindl（shrub species） had greater retention efficiency than Broussonetia papyrifera（broadleaf species） and Pinus bungeana Zucc.（coniferous species）, in terms of particle number and particle area cover.（2） 97.9% of the identified particles had diameter ≤10 μm, and 67.1% of them had diameter ≤2.5 μm. 89.8% of the particles had smooth boundaries, with 23.4% of them being nearly spherical.（3） 32.4%–74.1% of the particles were generated from bare soil and construction activities, and 15.5%–23.0% were mainly from vehicle exhaust and cooking fumes.