A comparison of supervised maximum likelihood and decision tree classification for crop cover estimation from multitemporal LANDSAT MSS data

Abstract

Supervised maximum likelihood classification was compared with a supervised binary decision tree for crop classification from multitemporal LANDSAT MSS data. Similar levels of classification accuracy were obtained using both algorithms, but the ease of training and computational simplicity of the binary decision tree suggest that this algorithm may be a viable alternative to the maximum likelihood for the analysis of data sets with high dimensionality such as multitemporal LANDSAT MSS data.     

Identification and area estimation of agricultural crops by computer classification of LANDSAT MSS data

LANDSAT Multispectral Scanner (MSS) data covering a three-county area in northern Illinois were classified using computer-aided techniques as corn, soybeans, or “other.” Recognition of test fields was 80% accurate. County estimates of the area of corn and soybeans agreed closely with those made by the USDA. Results of the use of a priori information in classification, techniques to produce unbiased area estimates, and the use of temporal and spatial features for classification are discussed. The extendability, variability, and size of training sets, wavelength band selection, and spectral characteristics of crops were also investigated.     

Atmospheric correction of LANDSAT MSS data for a multidate suspended sediment algorithm

This paper describes the application of various processing methods lo a synoptic set of satellite-sea data obtained from the Solent area on the south coast of England. The methods include ‘darkest pixel’ correction, sun angle and radiometric corrections, chromaticity analysis, atmospheric and surface corrections and the use of the satellite sensor response curves. It is shown that the use of a simple atmospheric/surface correction algorithm, based on atmospheric optical transmission theory, provides the most accurate method of estimating estuarine suspended sediment concentration from the satellite data above.     

Bright spot analysis of ocean-dump plumes using LANDSAT MSS

Identifying ocean-dumped materials by analysing the upwelled solar energy from the plume is complicated by the dispersion of the plume and the spectral absorption of the water. It is shown that the spectral analysis of ocean-dump plumes, using LANDSAT multispectral scanner (MSS) data, should be confined to the brightest area within the plume, the region where the waste material is least dispersed and nearest the surface. The decay of the upwelled radiance with time of the brightest pixel within the plume, at least for iron acid waste, is predictable. An accurate age determination of an acid plume is limited by striping within the MSS data     

A semi-automatic technique for multitemporal classification of a given crop within a landsat scene

A classification scheme based on temporal characteristics of a given crop is described. The technique in its present form requires one training field representative of the crop under consideration. This training field is used to determine analytically the time behavior of the crop in the LACIE (Large Area Crop Inventory Experiment) segment. A comparison of this crop's temporal profile, generated in each of the Landsat channels, with that of every pixel in the segment is made to decide the category (crop/noncrop) of the pixel. Classification results have been compared with ground truth for 34 sites in the U.S. Corn Belt. This technique has the potential for a more automated method of generating a near-harvest crop inventory from the satellite data in comparison to the inventory method in current use.     

An analysis of geologic lineaments seen on LANDSAT MSS imagery

Lineament maps drawn from several LANDSAT images of a part of north Wales and western England display considerable variation in the number of lineaments identified. Analysis of the maps shows that it is not the case that maps with fewer lineaments are simply subsets of those with many lineaments. Rather, each map contains a high proportion of lineaments that are unique to it. Despite these differences, the same preferred lineament orientation is identified from almost all maps. These results imply that all available LANDSAT imagery may usefully contribute to a lineament analysis, little value may be placed on the density of lineaments seen on any one image and preferred lineament orientation is relatively easy to identify. It is concluded that guarded use may be made of lineament analysis in geology. Lineament maps may be employed to suggest hypotheses rather than to test them.     

The saturation of LANDSAT MSS detectors over large ice masses

LANDSAT provides synoptic imagery for the study of large ice masses in the inaccessible polar regions. Even minor ice surface topographic detail can be identified as differences in relative brightness. However, radiance from such surfaces can be greater than the maximum MSS detector calibration. This results in detector saturation and consequent loss of information. Using MSS digital data from snow surfaces in both polar regions, corrected to radiance values to account for detector calibration changes, a model relating detector saturation in each MSS band to changes in Sun elevation is presented. Band 5 becomes saturated at the lowest Sun angles whereas band 7 remains unsaturated in all imagery examined. Secondary factors affecting detector saturation are atmospheric haze, the magnitude and orientation of surface slopes, variations in detector calibration and the character of the surface. Given a knowledge of the Sun elevation at which each band becomes saturated over large ice masses (of low gradient and snow-covered surface), and information on variations in Sun elevation with latitude and time of year, the spatial and temporal pattern of detector saturation can be predicted. This information is useful in the selection of MSS images for future glaciological studies, and in the scheduling by NASA of the optimum periods to acquire unsaturated MSS data of the polar regions.     

Effectiveness of user testing and heuristic evaluation as a function of performance classification

For different levels of user performance, different types of information are processed and users will make different types of errors. Based on the error's immediate cause and the information being processed, usability problems can be classified into three categories. They are usability problems associated with skill-based, rule-based and knowledge-based levels of performance. In this paper, a user interface for a Web-based software program was evaluated with two usability evaluation methods, user testing and heuristic evaluation. The experiment discovered that the heuristic evaluation with human factor experts is more effective than user testing in identifying usability problems associated with skill-based and rule-based levels of performance. User testing is more effective than heuristic evaluation in finding usability problems associated with the knowledge-based level of performance. The practical application of this research is also discussed in the paper.     

Derivation of atmospheric correction procedures for LANDSAT MSS with particular reference to urban data

The complexities and detail of urban scenes make it imperative that atmospheric effects are removed from satellite remotely sensed data prior to analysis. A study of atmospheric theory allows a simplified procedure to be developed for correction of multitemporal LANDSAT MSS scenes. Examples in the visible and near infrared, from a summer and winter scene, illustrate how disparate count values can be brought to good agreement as percentage reflectance.     

Contextual classification post-processing of LANDSAT data using a probabilistic relaxation model

A probabilistic relaxation model is used to improve maximum likelihood classifications of LANDSAT data of arid and urban areas in and around Ai Jahra, Kuwait. The problems of urban pixels, the role of compatibility coefficients and the iterations of the model are presented and discussed.     

Multitemporal land-cover classification of Mexico using Landsat MSS imagery

A complete land-cover classification of Mexico was performed using Landsat Multi-Spectral Scanner (MSS) imagery corresponding to years 1974, 1986 and 1990 ( - 1 y). The categorization of the approximately 2 M km 2 geographical region included the classification of approximately 300 equivalent scene images. Vegetation experts throughout the country provided an initial 250-class inventory of major vegetation associations by applying an unsupervised classification approach. A final regrouping was performed to produce a generalized thematic product containing 12 classes to provide a consistent national scale product. Classification accuracies were evaluated for each scene by means of cartographic comparison using two independently developed reference datasets corresponding to the 1970s and 1990s. An automated evaluation procedure was developed that incorporated decision rules to duplicate the results obtained using a manual accuracy assessment procedure. Overlaying both the image and the digital cartographic information allowed for the comparison of randomly selected pixels within each image scene. An overall accuracy for the three epochs of 62% was obtained for the 300 image scenes. Study results have provided a historical baseline documenting vegetation extent and distribution across Mexico over the two-decade period. This study serves as a possible model for subsequent North American land-cover characterization efforts.     

Classifier performance as a function of distributional complexity

When choosing a classification rule, it is important to take into account the amount of sample data available. This paper examines the performances of classifiers of differing complexities in relation to the complexity of feature-label distributions in the case of small samples. We define the distributional complexity of a feature-label distribution to be the minimal number of hyperplanes necessary to achieve the Bayes classifier if the Bayes classifier is achievable by a finite number of hyperplanes, and infinity otherwise. Our approach is to choose a model and compare classifier efficiencies for various sample sizes and distributional complexities. Simulation results are obtained by generating data based on the model and the distributional complexities. A linear support vector machine (SVM) is considered, along with several nonlinear classifiers. For the most part, we see that there is little improvement when one uses a complex classifier instead of a linear SVM. For higher levels of distributional complexity, the linear classifier degrades, but so do the more complex classifiers owing to insufficient training data. Hence, if one were to obtain a good result with a more complex classifier, it is most likely that the distributional complexity is low and there is no gain over using a linear classifier. Hence, under the model, it is generally impossible to claim that use of the nonlinear classifier is beneficial. In essence, the sample sizes are too small to take advantage of the added complexity. An exception to this observation is the behavior of the three-nearest-neighbor (3NN) classifier in the case of two variables (but not three) when there is very little overlap between the label distributions and the sample size is not too small. With a sample size of 60, the 3NN classifier performs close to the Bayes classifier, even for high levels of distributional complexity. Consequently, if one uses the 3NN classifier with two variables and obtains a low error, then the distributional complexity might be large and, if such is the case, there is a significant gain over using a linear classifier.     

Forest-type stratification and delineation of shifting cultivation areas in the eastern part of Arunachal Pradesh using LANDSAT MSS data

The Tirap district of Arunachal Pradesh has been mapped using LANDSAT data and reconnaissance-level forest-type maps have been prepared using visual and computer-aided techniques. A comparative study of two interpretation techniques has been carried out. The results of the study reveal that satellite remote sensing can be used for broad forest-type mapping and for monitoring degradation processes in the north-eastern region of the country; in this region resources information is limited and. due to fast changes, the information collected by conventional methods becomes outdated.     

Knowledge-based crop classification of a Landsat Thematic Mapper image

Abstract

A knowledge-based classification method was designed to improve crop classification accuracy. Crop data of preceding years, stored in a geographical information system (GIS) were used as ancillary data. Knowledge about crop succession, determined from crop rotation schemes, was formalized by means of transition matrices. The spectral data, the data from the GIS and the knowledge represented in the transition matrix were used in a modified Bayesian classification algorithm. The developed classification was tested in an agricultural region in The Netherlands. Depending on the spectral class discrimination, the accuracy of the knowledge-based classification was 6 to 20 percent better compared with a maximum likelihood classification.     

HSRS-SC:面向遥感场景分类的高光谱图像数据集

目的 场景分类是遥感领域一项重要的研究课题，但大都面向高分辨率遥感影像。高分辨率影像光谱信息少，故场景鉴别能力受限。而高光谱影像包含更丰富的光谱信息，具有强大的地物鉴别能力，但目前仍缺少针对场景级图像分类的高光谱数据集。为了给高光谱场景理解提供数据支撑，本文构建了面向场景分类的高光谱遥感图像数据集（hyperspectral remote sensing dataset for scene classification，HSRS-SC）。方法 HSRS-SC来自黑河生态水文遥感试验航空数据，是目前已知最大的高光谱场景分类数据集，经由定标系数校正、大气校正等处理形成。HSRS-SC分为5个类别，共1 385幅图像，且空间分辨率较高（1 m），波长范围广（380~1 050 nm），同时蕴含地物丰富的空间和光谱信息。结果 为提供基准结果，使用AlexNet、VGGNet-16、GoogLeNet在3种方案下组织实验。方案1仅利用可见光波段提取场景特征。方案2和方案3分别以加和、级联的形式融合可见光与近红外波段信息。结果表明有效利用高光谱影像不同波段信息有利于提高分类性能，最高分类精度达到93.20%。为进一步探索高光谱场景的优势，开展了图像全谱段场景分类实验。在两种训练样本下，高光谱场景相比RGB图像均取得较高的精度优势。结论 HSRS-SC可以反映详实的地物信息，能够为场景语义理解提供良好的数据支持。本文仅利用可见光和近红外部分波段信息，高光谱场景丰富的光谱信息尚未得到充分挖掘。后续可在HSRS-SC开展高光谱场景特征学习及分类研究。     

A scene image classification technique for a ubiquitous visual surveillance system

Multimedia Tools and Applications - The concept of smart cities has quickly evolved to improve the quality of life and provide public safety. Smart cities mitigate harmful environmental impacts and...     

An image recognition method by rough classification for a scene image

We have recognized the regions of scene images for image recognition. First, the proposed segmentation method classifies images into several segments without using the Euclidian distance. We need several features to recognize regions. However, they are different for chromatic and achromatic colors. The regions are divided into three categories (black, achromatic, and chromatic). In this article, we focus on the achromatic category. The averages of the intensity and the fractal dimension features of the regions in the achromatic category are calculated. We recognize the achromatic region by using a neural network with suitable features. In order to show the effectiveness of the proposed method, we have recognized the regions. This work was presented in part at the 10th International Symposium on Artificial Life and Robotics, Oita, Japan, February 4–6, 2005     

Using convolutional features and a sparse autoencoder for land-use scene classification

In this article, we propose a novel approach based on convolutional features and sparse autoencoder (AE) for scene-level land-use (LU) classification. This approach starts by generating an initial feature representation of the scenes under analysis from a deep convolutional neural network (CNN) pre-learned on a large amount of labelled data from an auxiliary domain. Then these convolutional features are fed as input to a sparse AE for learning a new suitable representation in an unsupervised manner. After this pre-training phase, we propose two different scenarios for building the classification system. In the first scenario, we add a softmax layer on the top of the AE encoding layer and then fine-tune the resulting network in a supervised manner using the target training images available at hand. Then we classify the test images based on the posterior probabilities provided by the softmax layer. In the second scenario, we view the classification problem from a reconstruction perspective. To this end we train several class-specific AEs (i.e. one AE per class) and then classify the test images based on the reconstruction error. Experimental results conducted on the University of California (UC) Merced and Banja-Luka LU public data sets confirm the superiority of the proposed approach compared to state-of-the-art methods.