Top PDF Shadow Detection in Aerial Images using Machine Learning

Shadow Detection in Aerial Images using Machine Learning

Shadow Detection in Aerial Images using Machine Learning

Texture-based methods exploit the fact that texture is retained in shadowed regions. Texture- based methods generally are implemented in two steps: 1) Finding candidate shadow pixels or regions, and 2) classifying them into either foreground or shadows. In the first step, shad- ows are selected based on spectral features. Then each shadow candidate region is classified as an object or shadow by checking the texture. If two regions have di ff erent spectral char- acteristics but the same texture information, then we choose the region with lower intensity as the shadowed region. Di ff erent types of correlation techniques are proposed (e.g. Gabor filtering [1]), Markov or conditional random fields [2, 3], orthogonal transforms [4], gradi- ent or edge correlation [5, 6, 7], normalised cross-correlation [8]). Texture-based methods can be powerful because textures are highly distinctive. Also, these methods are robust to change in illumination. However, texture-based methods are typically slow as they com- pute several neighborhood comparisons for each pixel. Hence, we did not explore this method as we are looking for methods that have faster run times.
Show more

83 Read more

Shadow detection in color aerial images based on HSI space and color attenuation relationship

Shadow detection in color aerial images based on HSI space and color attenuation relationship

This article is devoted to the problem of shadow detection in color aerial images. Hue singularity pixels are extracted. The candidate shadow and nonshadow regions are con- structed on the base of the modified ratio maps by using the Otsu ’ s thresholding method and the connected compo- nent analysis. The intensity property, chromaticity property of the shadow areas, and the color attenuation relationship derived from Planck ’ s blackbody irradiance law are used iteratively to segment each candidate region into smaller sub-regions, so that whether each sub-region is true shadow region is identified. The extracted hue singularity pixels are classified on the base of its neighboring pixels. From the above experimental results, it could be concluded that our proposed shadow detection algorithm presents best shadow detection accuracy when compared with Tsai ’ s and Chung et al. ’ s algorithms. Future work need to be done to solve the auto thresholds selection problem.
Show more

13 Read more

Towards an automatic road lane marks extraction based on ISODATA segmentation and shadow detection from large-scale aerial images

Towards an automatic road lane marks extraction based on ISODATA segmentation and shadow detection from large-scale aerial images

The preprocessing is designed to geometrically correct the raw aerial image and improve its quality. Raw digital images usually contain significant geometric distortions introduced by factors such as variations of the sensor platform, relief displacement, and nonlinearities in the sweep of a sensor’s IFOV, which makes it impossible for these images to be used directly as a map base without subsequent processing. Therefore, the geometric correction is employed to compensate for the distortions. The traditional photogrammetric triangulation process can be applied to the sequence of aerial digital images to generate ortho images under a predefined spatial coordinate system, which can be easily accomplished with the standard commercial digital photogrammetric software, i.e. ERDAR Leica Photogrammetry Suite (LPS). After the image ortho-rectification step, the distances between matched road lane markings can be accurately measured, and therefore the geometric specification of the lane marks can be utilized in the processing of road marking extraction.
Show more

11 Read more

Object Shadow Detection and Removal from Remotely Sensed Images

Object Shadow Detection and Removal from Remotely Sensed Images

In this section, our new analysis (algorithm) method is presented to detect shadows for color aerial images. The flowchart of our new analysis (algorithm) method is shown in Fig.2. We first transform RGB color image into the gray image and then, global thresholding (Otsu’s method) is used to determine a global threshold T for gray image to constructing the coarse-shadow map. Next to that morphology erosion operator with 5× 5 square structuring elements is applied.

11 Read more

Shadow Detection and Reconstruction in Satellite Images using Support Vector Machine and Image In-painting

Shadow Detection and Reconstruction in Satellite Images using Support Vector Machine and Image In-painting

There are various methods for shadow detection based on invariant colour features, physical properties of black body radiator etc [1]. Amani Massalabi proposed detecting information under and from shadow in panchromatic Ikonos images of the city of Sherbrooke [2]. Many techniques for shadow detection are developed for video images based on color properties and temporal frame difference. Elena Salvador proposed shadow identification and classification using invariant color models [3]. This method requires luminance and color information for shadow identification. Pooya Sarabandi proposed a method for shadow detection and Radiometric Restoration in Satellite high resolution images [4]. This method detects the boundaries of cast shadow in high resolution satellite images. Also, the shadow detection is performed on the basis of radiometric techniques such as gamma correction, linear-correlation and histogram matching. Kuo-Liang Chung proposed efficient shadow detection of color aerial images based on successive thresholding scheme [5].
Show more

8 Read more

Analysis of Mammographic Images for Early Detection of Breast Cancer Using Machine Learning Techniques

Analysis of Mammographic Images for Early Detection of Breast Cancer Using Machine Learning Techniques

The most important thing about estimation of mammography is that it can identify breast variations in the early stages of cancer even before the development any physical symptoms. The American Cancer Society's rules for early breast malignancy location stress mammography and physical examinations. Clearly there are numerous different methods and techniques that are utilized for breast screening and every strategy accomplishes an alternate level of clarity in showing breast images. On the other hand, mammography is the main procedure that has been ended up being successful for breast tumour screening. One of the primary favourable circumstances of utilizing mammography is its cheap expense of usage for a huge populace of subjects. Since overall radiologists screen more than hundreds of films each day, keeping up consistency and exactness in analysis is not simple. This implies that computer supported analytic systems have the best seek after enhancing breast disease identification and reducing morbidity from the disease.
Show more

52 Read more

Detection and Removal of Shadow Using Very High Resolution Remote Sensing Images

Detection and Removal of Shadow Using Very High Resolution Remote Sensing Images

ABSTRACT- The shadows are mainly observed because of tall buildings, towers etc. in urban areas. Shadows in very high resolution (VHR) Remote Sensing images represent serious problems for their full development. To reduce the shadow effects in very high resolution (VHR) Remote Sensing images for their further applications, detection and removal of shadow is necessary which is developed. In this project we have addressed the issue of shadow detection and removal in VHR Remote Sensing images.The detection and classification tasks are implemented by means of support vector machine approach and for noise removal filtering is used. Shadow removal is done using linear regression method. In this method, the shadow blocks of the image are replaced by adjusting the intensities of the shaded points to the statistical characteristic of the non-shadow regions. Then fuse the original image and shadow removal image. Wavelet transform is used for image fusion. Inverse wavelet transform is used for getting original image back from decomposed image.
Show more

7 Read more

Road detection and segmentation by aerial images using CNN based system

Road detection and segmentation by aerial images using CNN based system

The Extraction of reliable information from aerial images is difficult problem, but it has numerous important utilizations: The disaster monitoring crop monitoring in precision agriculture, border surveillance, traffic monitoring, and so on. different image processing techniques were considered. Texture analysis techniques are used to detect and segment regions of interest and, particularly roads, from aerial images in and but the choice of the representative features depends on the specific context of the application that uses it. The authors in consider also a supervised learning approach to detect road textures using a neural network. UAV (multi- copter type) is proposed for efficient road detection and tracking. Different road features and information as the Stroke Width Transform, colors, and width, are combined to highlight possible road candidates [6]. In order to increase the accuracy and robustness of road detection in a deep Convolutional Neural
Show more

5 Read more

CRACK DAMAGE DETECTION IN UNMANNED AERIAL VEHICLE IMAGES OF CIVIL INFRASTRUCTURE USING PRE-TRAINED DEEP LEARNING MODEL

CRACK DAMAGE DETECTION IN UNMANNED AERIAL VEHICLE IMAGES OF CIVIL INFRASTRUCTURE USING PRE-TRAINED DEEP LEARNING MODEL

DCNNs typically require large annotated image datasets to achieve high predictive accuracy. However, in many domains, acquisition of such data is difficult and labeling them is costly. In light of these challenges, the use of ‘off-the-shelf’ DCNN features of well-established DCNNs such as VGG-16, AlexNet, and GoogLeNet pre- trained on large-scale annotated natural image datasets (such as ImageNet) have been shown to be very useful for solving cross domain image classification problems through the concept of transfer learning and fine-tuning (Shin et al., 2016). In DCNN, representations learned at different layers of the network correspond to different levels of abstractions present in the input images. The initial layers extract edges and color information while the latter layer filters encode shape and texture. The idea behind transfer learning is that it is cheaper and efficient to use deep learning models trained on “big data” image datasets (like
Show more

14 Read more

Survey on Cancer Detection using Machine Learning

Survey on Cancer Detection using Machine Learning

be a noninvasive skin imaging process that enhances the visual effect of a skin lesion by removing surface reflection to gain a magnified and luminous image of a skin region for greater clarity of points [2]. However, it remains a difficult task to recognize automatically the melanoma in dermoscopic images because of its many challenges. First, it is hard to classify correct lesion areas because of the poor contrast between skin lesions and normal skin layer. Second, a high level of visual resemblance may exist in the lesions of Melanoma and non-Melanoma, contributing to differentiating between the lesions of Non-Melanoma melanoma. Shift between individuals in the presentation of melanoma, e.g. colour and shape, natural hair or veins, and others; thirdly, the consistency in skin conditions. Segmentation of skin lesions is the key to many approaches to classification. There is often a recent review of automatic algorithms for segmentation of skin lesions. The accuracy of the following lesion classification may benefit from accurate segmentation.
Show more

7 Read more

A System Which Supports Shadow Detection and Shadow Removal for High Resolution Remote Sensing Images

A System Which Supports Shadow Detection and Shadow Removal for High Resolution Remote Sensing Images

Shadow will occur by sunlight or any light sources. We cannot get clear and quality picture for obtain the shadow in the images. The objective of this paper is detected and removal of shadows plays an important role in application of urban high resolution remote sensing images. Object oriented shadow detection and removal methods are used in this paper. Shadow detection is used during the image segmentation. The suspected shadows are extracted by the statistical features. For shadow removal, support vector machine and adaboost classifier based on IOOPL matching could effectively remove the shadow. According to the homogeneous section shadow removal will be performed. Our method can accurately detect shadows from urban high- resolution remote sensing images and can effectively restore shadows with a rate of over 95%.
Show more

6 Read more

A NOVEL SHADOW DETECTION AND SHADOW REMOVAL ALGORITHM FOR OPTICAL SATELLITE IMAGES

A NOVEL SHADOW DETECTION AND SHADOW REMOVAL ALGORITHM FOR OPTICAL SATELLITE IMAGES

Therefore, it is necessary to reduce or to compensate the effect of shadows in order to get the fine information. Occurrence of shadows reduce the successful rate of various methods like edge extraction, object recognition, and image matching and change detection for the corresponding ground objects in the shadow [12, 14]. Several literatures have useful information about shape, relative position, surface character and other characters of object generating shadow [1,6,17]. Shadow detection and removal methods work together to remove shadows. They presented the technique for both VHSR and aerial satellite images and it can be categorized into method based on model and method based on shadow property [15][5][2][12][16].Spectral and geometrical features are used in shadow property based methods [4].Tsai gives a method using spectral images in HSI space to segment shadow [11] and also a method using thresholding saturation intensity difference in HSI color space [7]. After shadow detection, shadow removal can be done to eliminate the shadows and to restore the regions under shadow before using in any applications. Shadow compensation technique is used to compensate the shadow pixels and to restore the regions under shadows. Several algorithms include gamma correction method, linear correlation method, posteriori probabilities method [3,8,9] proposed that they are operate band by band in RGB color space, which
Show more

8 Read more

Crowd detection from aerial images

Crowd detection from aerial images

Chapter three proposes the methodology. In this chapter the methods and steps that are proposed in order to perform crowd detection are described. The methods consist of; Features from Accelerated Segment Test (FAST) feature extraction, further features extraction using Gray Level Co-occurrence Matrices (GLCM) and Support Vector Machine (SVM) to classify between crowd and non- crowd.

24 Read more

Mid to Late Season Weed Detection in Soybean Production Fields Using Unmanned Aerial Vehicle and Machine Learning

Mid to Late Season Weed Detection in Soybean Production Fields Using Unmanned Aerial Vehicle and Machine Learning

background but were difficult to differentiate between the crops and the weeds. Texture features that capture the spatial variation of pixel intensities as well as shape features such as roundness, perimeter among others, were able to distinguish between broadleaf and grassy plants. However, they were not able to differentiate individual species of weeds (D. M. Woebbecke, G. E. Meyer, K. Von Bargen, & D. A. Mortensen, 1995a, 1995b; G. E. Meyer, T. Mehta, M. F. Kocher, D. A. Mortensen, & A. Samal, 1998). Following this, with advancements in variable rate implements, several studies focused on site-specific spraying. These systems used image processing techniques like discrete wavelet transform as well as nonlinear classifiers to recognize the weeds. Besides, these systems used the information about crop row to detect the inter-row weeds. However, the speed of these systems was limited by the computational capacity of the hardware (L. Tian, J. F. Reid, & J. W. Hummel, 1999; W. S. Lee, Slaughter, & Giles, 1999). Thorp & Tian (2004) studied the potential of satellite and manned aircraft-based remote sensing technologies to locate the weed patches. Using aerial imagery enables sensing large areas to develop a prescription map for herbicide application which can then be used by
Show more

125 Read more

Detection of Cystic Fibrosis Symptoms Based on X-Ray
Images Using Machine Learning- Pilot Study

Detection of Cystic Fibrosis Symptoms Based on X-Ray Images Using Machine Learning- Pilot Study

Machine Learning (ML) is a science dealing with algorithms and statistical models that allow computer systems to perform certain tasks without giving exact instructions - they only use emerging patterns and dependencies. They are used in those areas for which the creation of a working and correct algorithm solving a given problem is very difficult, and sometimes even impossible. Artificial Neural Networks (ANN) are one type of machine learning algorithms [9]. These are systems of elements (nodes) loosely inspired by biological neural networks. They work like synapses in the human brain - training data causes signals to be sent along a network - according to the appropriate mathematical function (usually non-linear) assigned to each of the neurons. ANN requires a large amount of “training” data to calculate the weights of nodes of the network [10]. From several to thousands of neurons can make up a single layer, each of which can serve different purposes and perform different operations on the input data. The basic algorithm that teaches neural networks is the backpropagation algorithm, which involves finding the weights of individual neurons moving from the final (output) layer through subsequent hidden layers to the input layer.
Show more

6 Read more

Detection of Tree Crowns Based on Reclassification Using Aerial Images and Lidar Data

Detection of Tree Crowns Based on Reclassification Using Aerial Images and Lidar Data

Detection and classification of objects on earth were and still are important fields for researchers in different majors including Remote Sensing and Photogrammetry (Rottensteiner et el., 2011). As emerging new sensors like laser scanners, developing Photogrammetry field, and utilizing digital cameras, methods of detection and classification are got into new era. High resolution and high spectral digital cameras have lead researchers to develop and introduce new indices to detect a variety of objects on earth. LiDAR data give 3D coordinates of points directly, that this ability makes it a simple task to differentiate between smooth and rough planes. Smooth planes usually designate man-made objects and rough planes mostly designate natural grounds. Because LiDAR is an active sensor it has no problem dealing with shadow areas, while shadow areas are challenging in aerial images. At high resolution aerial images boundary of objects like Buildings is clearly notable, but in LiDAR data there are some problems in detecting such boundaries. Considering advantages and disadvantages both LiDAR and aerial imagery it seems integrating these two data sources is the best option (Rottensteiner et el., 2008). Tree detection in complex city scenes because of existing high buildings is a more difficult task than tree detection in plains and cities with low buildings. There are lots of methods and algorithms in detection and classification field but it is not possible to compare those together. This is because of lack of bench mark data sets. In other hand most of algorithms and methods have tested in different data sets. To overcome this problem and making it easier to compare methods together a
Show more

6 Read more

Detection and Analysis Methods for unmanned aerial Vehicle Images

Detection and Analysis Methods for unmanned aerial Vehicle Images

Abstract – This chapter presents innovative methods for the automatic detection and counting of cars and for the estimation of their speed in Unmanned Aerial Vehicle (UAV) images acquired over urban contexts. UAV images are characterized by an extremely high spatial resolution, which makes the detection of cars particularly challenging. The proposed car detection and counting methods start with a screening operation in which the asphalted areas are identified in order to make the processes faster and more robust. Then, after the screening operation, both methods proceed with their respective core part. The first method performs a feature extraction process based on Scalar Invariant Feature Transform (SIFT) thanks to which a set of keypoints is identified and opportunely described. Successively, it discriminates between keypoints assigned to cars and all the others, by means of a Support Vector Machine (SVM) classifier. Differently, the second method carries out filtering operations in the horizontal and vertical directions to extract Histogram of Gradient (HoG) features and to yield a preliminary detection of cars after the computation of a similarity measure with a catalogue of cars used as reference. Three different strategies for computing the similarity are investigated. Successively, for the image points identified as potential cars, an orientation value is computed by searching for the highest similarity value in 36 possible directions. In the end, both methods group the points (i.e., the car keypoints or the image points associated to the car class) belonging to the same car in order to get a “one point – one car” relationship thanks to a spatial clustering. Furthermore, in this chapter, a method to monitor the traffic by tracking and estimating the speed of moving cars from sequences of images is described. The method begins with the automatic registration of pairs of consecutive images of a sequence by using a geometric transformation obtained from the matching of invariant points. By comparing the images and thanks to mathematical morphology operations the moving cars are isolated. In the last step, the spatial coordinates in both images of each car are extracted and thanks to the derived spatial shift estimations about the speeds are inferred. Interesting experimental results obtained with all the proposed strategies and conducted on sets of real UAV images are presented and discussed.
Show more

108 Read more

Building Detection using Aerial Images and Digital Surface Models

Building Detection using Aerial Images and Digital Surface Models

In this paper a method for building detection in aerial images based on variational inference of logistic regression is proposed. It consists of three steps. In order to characterize the appearances of buildings in aerial images, an effective bag-of-Words (BoW) method is applied for feature extraction in the first step. In the second step, a classifier of logistic regression is learned using these local features. The logistic regression can be trained using different methods. In this paper we adopt a fully Bayesian treatment for learning the classifier, which has a number of obvious advantages over other learning methods. Due to the presence of hyper prior in the probabilistic model of logistic regression, approximate inference methods have to be applied for prediction. In order to speed up the inference, a variational inference method based on mean field instead of stochastic approximation such as Markov Chain Monte Carlo is applied. After the prediction, a probabilistic map is obtained. In the third step, a fully connected conditional random field model is formulated and the probabilistic map is used as the data term in the model. A mean field inference is utilized in order to obtain a binary building mask. A benchmark data set consisting of aerial images and digital surfaced model (DSM) released by ISPRS for 2D semantic labeling is used for performance evaluation. The results demonstrate the effectiveness of the proposed method.
Show more

7 Read more

Object Detection in High Resolution Aerial Images and Hyperspectral Remote Sensing Images

Object Detection in High Resolution Aerial Images and Hyperspectral Remote Sensing Images

Over the past few years, CNNs have become the mainstream approaches in the computer vision field and achieved breakthrough performances in many tasks, including image clas- sification [21] and object detection [135]. In the literature, the CNN based detection algorithms are divided into two categories [135]: the two-stage region-based framework (R-CNN [136], Fast R-CNN [137], and Faster R-CNN[25]) and the one-stage unified ap- proach (OverFeat [138], SSD [26], YOLO [139, 27]). Compared to the traditional object detection algorithms, R-CNN was among the early attempts to apply CNNs for object detection. The algorithm first applied the selective search approach [140] to generate a large number (around 2000) of object candidate regions (region proposals). Then, a pre- trained CNN network (AlexNet [21]) was utilized to extract a 4096-dimensional feature vector for each region proposal. Following that, a linear support vector machine (SVM) was trained to determine the object categories. To improve the efficiency in generating feature vectors for all region proposals, Fast R-CNN [137] was proposed. It extends the R-CNN approach by sharing the computations across region proposals and incorporating the ROI pooling layer to get the fixed-length feature representations for image regions. To further improve the detection speed, faster R-CNN [25] proposed the region proposal network (RPN), which learns to generate region proposals and completely eliminates the need for getting pre-defined candidate regions. Specifically, two separate network branches were attached to the base network to generate region proposals and object classification results.
Show more

137 Read more

Automated elephant detection and classification from aerial infrared and colour images using deep learning

Automated elephant detection and classification from aerial infrared and colour images using deep learning

the layer weights that need to be tuned during training to extract and activate on relevant features in the input image. Each convolutional layer in a network can flow into another convolutional layer, which allows the network to learn hierarchical features [13]. Starting with simple image parts such as edges and gradients in the first layers, we end up with complex object parts such as elephant body shapes. An example of a single filter applied to our elephant images is shown in Figure 2.6. It seems to successfully extract the shape information of an elephant (amongst other shapes). The size of a convolutional kernel is called the receptive field. By learning filter weights that move across an input image we are locally connecting our filter weights to the input. This reduces the number of parameters we need to train (called parameter sharing), and allows us to train features that are translation invariant. We specify the depth of a convolutional layer to determine the number of unique features we would like the layer to learn, and this depth is a hy- perparameter of the layer. Another hyperparameter is the stride length, which specifies by how many pixels the filter will move between every convolution operation. A larger stride size will result in a smaller output (as well as more information loss between layers). Since a convolution operation results in information reduction we typically end up with a smaller output compared to our input. By zero-padding the output we can control its size as it moves through our network. Each convolutional filter output is called a feature map. All feature maps created by a convolu- tional layer pass through an activation function to produce an activation map [12], sometimes called the detector stage [13]. This is the same nonlinear activation function discussed earlier, with ReLU being a popular option due to its positive impact on training times. ReLU can be im- plemented as a simple threshold operation, compared to the more computationally expensive activations, such as the sigmoidal function.
Show more

80 Read more

Show all 10000 documents...