SDM College of Engineering and Technology

About: SDM College of Engineering and Technology is a based out in . It is known for research contribution in the topics: Diesel fuel & Combustion. The organization has 350 authors who have published 351 publications receiving 2399 citations.

An improved technique of plant leaf classificaion using hybrid feature modeling

Abstract: Various applications have been developed during recent years which are based on the computer vision system. In this field, plant species recognition is a challenging task for researchers due to environmental and image acquisition condition of image. Leaf classification application can be used for various purpose such as remote sensing imaging, botanical characteristically analysis etc. Now a day, amount of dataset is increasing rapidly in this field. Thus it motivates us to develop an efficient model for leaf classification in terms of computation time and accuracy. To address this issue, here in this work we propose a hybrid approach for feature extraction to improve the classification accuracy. Proposed model follows twofold working process whereas in first stage image pre-processing is applied to enhance the image quality of image by applying image enhancement with the help of auto-regressive model, second stage performs feature computation by combining morphological, shape and SIFT feature and finally Deep Neural Network is applied for classification performance evaluation. Proposed work is a combination of image enhancement, morphological feature, SIFT feature and classification technique. For efficient image enhancement, auto-regressive model is adopted here. A comparative study of classification performance is presented which shows the robustness of proposed model

Adaptive neuro-fuzzy technique for performance tuning of database management systems

Abstract: A recent trend in database performance tuning is towards self tuning for some of the important benefits like efficient use of resources, improved performance and low cost of ownership that the auto-tuning offers. Most modern database management systems (DBMS) have introduced several dynamically tunable parameters that enable the implementation of self tuning systems. An appropriate mix of various tuning parameters results in significant performance enhancement either in terms of response time of the queries or the overall throughput. The choice and extent of tuning of the available tuning parameters must be based on the impact of these parameters on the performance and also on the amount and type of workload the DBMS is subjected to. The tedious task of manual tuning and also non-availability of expert database administrators (DBAs), it is desirable to have a self tuning database system that not only relieves the DBA of the tedious task of manual tuning, but it also eliminates the need for an expert DBA. Thus, it reduces the total cost of ownership of the entire software system. A self tuning system also adapts well to the dynamic workload changes and also user loads during peak hours ensuring acceptable application response times. In this paper, a novel technique that combines learning ability of the artificial neural network and the ability of the fuzzy system to deal with imprecise inputs are employed to estimate the extent of tuning required. Furthermore, the estimated values are moderated based on knowledgebase built using experimental findings. The experimental results show significant performance improvement as compared to built in self tuning feature of the DBMS.

Estimating distance threshold for greedy subspace clustering

Abstract: Many approaches have been proposed to recognize clusters in subspaces. However, their performance is highly sensitive to input parameter values. The purpose and expected ranges of these parameters may not available to a non-expert user. The parameter setting producing optimal results can only be known after repeated execution of the clustering process every time with a different set, which is very time consuming. Most of the existing algorithms show high runtimes due to excessive data scans. In this work, we propose a subspace clustering technique that estimates the distance threshold parameter automatically from the data for each attribute and works on the basis of single linkage clustering, in bottom up, greedy fashion. The experimental results show that, the algorithm produces optimal results without accepting any input from the user, achieves up to 10 times better runtime and improved accuracy in a single run without requiring any tuning of parameter values.

Comprehensive Analysis of Object Detection through Segmentation

Abstract: In computer vision extracting an object from an image automatically is too hard Towards addressing this issue a comprehensive analysis of most of the Object detection through different Segmentations is performed taken from the major recent publications covering various aspects of the research in this area We identify the following methods of the state-of-the-art techniques in which an object can be detected: (1) Mean Shift Segmentation With Region Merging, (2) Boundary Structure Segmentation With Region Grouping, (3) Watershed Segmentation With Region Merging All these are semi automatic detection of an object through segmentation and contour based shape descriptor The results tabulated prove that the Mean Shift Segmentation with Region Merging Process yields the best result over the other two methods in detection the Object Of Interest

Recognition of Traffic Sign Based on Support Vector Machine and Creation of the Indian Traffic Sign Recognition Benchmark

Abstract: Traffic sign recognition, a driver assistance system informs and warns the driver about the status of the road is a challenging issue. Though, a lot of work on this topic has been carried out, but complete benchmark datasets are not freely available for comparison of different approaches. A few databases are available for benchmarking automatic detection of traffic signs. However, there is no database built considering the Indian traffic signs. The road scenarios in India are quite different from other countries, especially in rural areas. Hence, an effort to build an Indian traffic sign database considering both rural and urban situations is presented in the work. The database consists of 13000 traffic sign images of 50 different classes of traffic signs taken at different times under different environmental conditions and includes the detailed annotation of the traffic signs in terms of size, type, orientation, illumination and occlusion. The work also discusses an efficient method for identification of road signs based on two modules: (1) feature extraction based on dense scale invariant feature transform (DSIFT) and (2) a classifier trained by support vector machine (SVM). The SIFT approach transforms an image it into a large collection of local feature vectors invariant to scaling, translation or rotation of the image, and reduction in the dimensionality is achieved by applying principal component analysis (PCA). After extracting the features, the image is classified using support vector machine, a supervised learning model.