Selection of features system and network parameters for hyperspectral images classification using convolutional neural networks

A classification system for hyperspectral images using convolutional neural networks is described. A specific network was selected and analyzed. The network parameters, ensured the maximum classification accuracy: dimension of the input layer, number of the layers, size of the fragments into which the classified image is divided, number of learning epochs, are experimentally determined. High percentages of correct classification were obtained with a large-format hyperspectral image, and some of the classes into which the image is divided are very close to each other and, accordingly, are difficult to distinguish by hyperspectra.

EELGRASS MAPPING WITH SENTINEL-2 AND UAV DATA IN PRINCE EDWARD ISLAND (CANADA)

Eelgrass (Zostera marina L.) is a marine angiosperm that grows throughout coastal regions in Atlantic Canada. Eelgrass beds provide a variety of important ecosystem services, and while it is considered an important marine species, little research has been done to understand its distribution and location within Atlantic Canada. The purpose of this study was to assess the capability of Sentinel-2 and UAV imagery to map the presence of eelgrass beds within the Souris River in Prince Edward Island. Both imageries were classified using the non-parametric Random Forests (RF) supervised classifier and the resulting classification was validated using sonar data. The Sentinel-2 classified image had a lower validation accuracy at 77.7%, while the UAV classified image had a validation accuracy of 90.9%. The limitations of the study and recommendations for future work are also presented.

ANALYSIS OF VHR IMAGE CLASSIFICATION BY SINGLE AND ENSEMBLE OF CLASSIFIERS

Given the wide variety of image classifiers available nowadays, some questions remain about the accuracy and processing time of Very High Resolution (VHR) images. Another question concerns the use of a Single or Ensemble Classifiers. Of course, the main factor to consider is the quality of the classified image, but computational cost is also important, especially in applications that require real-time processing. Given this scenario, this paper aims to relate the accuracy of seven single classifiers and the ensemble of the same classifiers with the processing time. In this paper the ensemble of classifiers had the best results in terms of accuracy, however, it comes to processing time, the decision tree had the best performance.

Wetland Mapping with Landsat 8 OLI, Sentinel-1, ALOS-1 PALSAR, and LiDAR Data in Southern New Brunswick, Canada

Mapping wetlands with high spatial and thematic accuracy is crucial for the management and monitoring of these important ecosystems. Wetland maps in New Brunswick (NB) have traditionally been produced by the visual interpretation of aerial photographs. In this study, we used an alternative method to produce a wetland map for southern New Brunswick, Canada, by classifying a combination of Landsat 8 OLI, ALOS-1 PALSAR, Sentinel-1, and LiDAR-derived topographic metrics with the Random Forests (RF) classifier. The images were acquired in three seasons (spring, summer, and fall) with different water levels and during leaf-off/on periods. The resulting map has eleven wetland classes (open bog, shrub bog, treed bog, open fen, shrub fen, freshwater marsh, coastal marsh, shrub marsh, shrub wetland, forested wetland, and aquatic bed) plus various non-wetland classes. We achieved an overall accuracy classification of 97.67%. We compared 951 in-situ validation sites to the classified image and both the 2106 and 2019 reference maps available through Service New Brunswick. Both reference maps were produced by photo-interpretation of RGB-NIR digital aerial photographs, but the 2019 NB reference also included information from LiDAR-derived surface and ecological metrics. Of these 951 sites, 94.95% were correctly identified on the classified image, while only 63.30% and 80.02% of these sites were correctly identified on the 2016 and 2019 NB reference maps, respectively. If only the 489 wetland validation sites were considered, 96.93% of the sites were correctly identified as a wetland on the classified image, while only 58.69% and 62.17% of the sites were correctly identified as a wetland on the 2016 and 2019 NB reference maps, respectively.

Land use land cover change detection in the mining areas of V.D. Yalevsky coal mine-Russia

The dynamics of land use/land cover (LULC) changes, the effect of coal mining on the LULC changes, and the regional environmental impact are discussed in this study. The different land use classes mainly Forest, Water Bodies, Road, Mining Area, Agriculture and Grass in the study area of V.D. Yalevsky coal field area in Kemerovo region of Russia are identified. On the other hand the impact of V.D. Yalevsky coal mine activities on LULC change on the environment and teritory are discussed. The LULC changes in the V.D. Yalevsky coal field area were analyzed for a period of 27 years e.g., from the year 1992 to 2019. The changes were detected on a 13-years time interval using Landsat-4 TM, Landsat-8 OLI. Furthermore supervised classification techniques using maximum likelihood method through ENVI (Environment for Visualizing Images) 5.1software was utilized. In addition post classification change detection method through ENVI was used to investigate the changes. The study reveals decrement in LULC categories of forest to 25.35km², water bodies to -0.94km², agriculture to -98.48km², road to -10.80km². However increment in the rate of mining area to 100.72km² and grass cover 34.86 km² during the study period. Meanwhile 90.18 % overall accuracy and (0.87) kappa coefficient for 1992 classified image, 93.41 % overall accuracy and (0.91)Kappa coefficient for 2006 classified image and 88.69 % overall accuracy and (0.85) kappa coefficient for 2019 classified image were obtained.

Sensor-based Waste Handling System

Earlier, separation of waste objects was a tedious process for humans since it requires thorough identification of each object’s nature. The identification and segregation of waste products are indispensable processes. The project consists of an Image Classification section where the waste is captured with the help of Raspberry pi camera and processed in the appropriate environment to classify if the waste is biodegradable or nonbiodegradable. The classified image is set with a key and delivered to the breadboard which is connected with Raspberry pi to illuminate the LED accordingly. The untrained or unidentified object is marked with a different LED and can be left for a new training process so that the system collects the features of the particular object and be ready with a model. Following is the Waste Management System. An Ultrasonic sensor is placed at the corner to dump the waste in the corresponding bin with the help of servo motor, which contributes to swap the bins by rotating itself in 180 degrees when non-biodegradable waste is identified. The classified object is disposed in its bin which concludes both the classification and segregation processes. Manual labour is minimized through this automatic waste identification and disposal

Object Detection and Tracking using YOLO v3 Framework for Increased Resolution Video

The proposed system is used for vehicle detection and tracking from the high-resolution video. It detects the object (vehicles) and recognizes the object comparing its features with the features of the objects stored in the database. If the features match, then object is tracked. There are two steps of implementation, online and offline process. In offline process the data in the form of images are given to feature extractor and then after to the trained YOLO v3 model and weight files is generated form the pre-trained YOLO v3 model. In online phase, real-time video is applied to feature extractor to extract the features and then applied to the pre-trained YOLO v3 model. The other reference to YOLO v3 model pre-trained is the output of weight file. The YOLO v3 model process on the video frame and weight file extracted features, the model output is classified image. In YOLO v3 Darknet-53 is used along with Keras, some libraries with OpenCV, Tensor Flow, and Numpy. The proposed system is implemented on PC Intel Pentium G500, 8GB and operating system Windows 7 is used for processing our system. The system is tested on PASCAL VOC dataset and the results obtained are accuracy 80%, precision 80%, recall 100%, F1-Score 88%, mAP 76.7%, and 0.018%. The system is implemented using python 3.6.0 software and also tested using real-time video having 1280x720 and 1920x1080 resolutions. The execution time for one frame of video having resolution of 1280x720 (HD) and 1920x1080 (FHD) and 1280x720 (HD) are 1.840 second and 4.414808 seconds respectively with accuracy is 80%.

Land use land cover change detection in the mining areas of V. D. Yalevsky coal mine-Russia

The dynamics of land use/land cover (LULC) changes, the effect of coal mining on the LULC changes, and the regional environmental impact are discussed in this study. The different land use classes mainly Forest, Water Bodies, Road, Mining Area, Agriculture and Grass in the study area of V. D. Yalevsky coal field area in Prokorvisk city in Kamerovo region of Russia are identified. On the other hand the impact of V. D. Yalevsky coal mine activities on LULC change on the environment and teritory are discussed. The LULC changes in the V. D. Yalevsky coal field area were analyzed for a period of 27 years e.g., from the year 1992 to 2019. The changes were detected on a 13-years time interval using Landsat-4 TM, Landsat-8 OLI. Furthermore supervised classification techniques using maximum likelihood method through ENVI (Environment for Visualizing Images) 5.1software was utilized. In addition post classification change detection method through ENVI was used to investigate the changes. The study reveals decrecment in LULC cotogories of forest to 25.35km², water bodies to -0.94km², agriculture to -98.48km², road to -10.80km². However increment in the rate of mining area to 100.72km² and grass cover 34.86 km² during the study period. Meanwhile 90.18% overall accuracy and (0.87) kappa coefitient for 1992 classified image, 93.41% overall accuracy and (0.91) Kappa koefitient for 2006 classified image and 88.69% overall accuracy and (0.85) kappa coefitient for 2019 classified image were obtained.

Classification of the galaxy Milky Way using variable precision rough sets technique

Astronomy image is regarded main source of information to discover outer space, therefore to know the basic contain for galaxy (Milky way), it was classified using Variable Precision Rough Sets technique to determine the different region within galaxy according different color in the image. From classified image we can determined the percentage for each class and then what is the percentage mean. In this technique a good classified image result and faster time required to done the classification process.

Studying the contribution of components and type of spiral galaxy NGC 6946 using digital image processing

NGC 6946 have been observed with BVRI filters, on October 15-18,2012, with the Newtonian focus of the 1.88m telescope, Kottamiaobservatory, of the National Research Institute of Astronomy andGeophysics, Egypt (NRIAG), then we combine the BVRI filters toobtain an astronomical image to the spiral galaxy NGC 6946 whichis regarded main source of information to discover the components ofthis galaxy, where galaxies are considered the essential element ofthe universe. To know the components of NGC 6946, we studied itwith the Variable Precision Rough Sets technique to determine thecontribution of the Bulge, disk, and arms of NGC 6946 according todifferent color in the image. From image we can determined thecontribution for each component and its percentage, then what is thepercentage mean. In this technique a good classified image resultand faster time required to done the classification process.