Mepco Schlenk Engineering College

About: Mepco Schlenk Engineering College is a based out in . It is known for research contribution in the topics: Wavelet & Wavelet transform. The organization has 1307 authors who have published 1665 publications receiving 18690 citations.

Classification of citrus and non-citrus fruits using texture features

Abstract: Each species of a fruit has its own unique features, which enabled the proposed system to identify it accurately. The proposed system classifies citrus from Non-citrus fruits using the textural features. Features are extracted based on the statistical and co-occurrence features derived from Wavelet transformed sub bands. The classification is done by the minimum distance criterion. Experimental results on a database of about 200 fruits confirm the effectiveness of the proposed approach.

Self Configurations, Optimization and Protection Scenarios with wireless sensor networks in IIoT

Abstract: Today’s Industrial Automation process requires more number of devices and sensor/actuators are communicating each other. It leads into more diversity. In this situation, the controlling part of the connected sensor devices and communication devices are a crucial one. The main purpose of the work is to reduce the latency problem and congestion. Hence, our focus is to design a smart sensor controller that provides reliable data transmission between process controller and sensor node by using Critical Communication and Massive IoT. In our paper, the data transfer is done by using various technology protocols, such as Bluetooth, Wi-Fi, GSM and Zigbee. The smart sensor element with the characteristics of SCOP (self-configuration, self-optimization and self- protection) is synthesized and implemented in FPGA Spartan 6 along link with ATMEGA328 microcontroller for exchange of protocols. It develops an analysis framework for IIoT that can be used to enumerate and characterize IIoT devices when studying system architectures.

Handwritten Text Segmentation Using Pixel Based Approach

Abstract: This paper presents a simple approach for the character segmentation of the handwritten words by bounding box approach and pixel based approach. The handwritten character segmentation is a tedious process because of their unconstrained writing styles. The handwritten words are classified based on their writing method. The non-touching words are segmented by the bounding box approach and the touching words are segmented by the pixel-wise approach. The novel part of this paper is to avoid over segmentation by thresholding automatically. The characters which are separated are then subjected to recognition using KNN, the non-parametric classifier. The features extracted for the classifying the model are Histogram of Oriented Gradients (HOG), Log-Gabor filters, concatenation of both the features and some geometric features. The segmentation accuracy is also responsible for the performance of the recognition. A benchmark database of IAM handwritten words is used for the handwritten segmentation and recognition. A random subset of words are taken into account and carried out for the proposed work. This paper achieves the segmentation rate of 94.45% and the recognition rate for 50-50% training and testing ratio is 85.89%.

Generalized Neural Network Model to Predict the Properties of Sintered Al - Fe Composite

Abstract: The deformation and strain hardening behaviour of Al-Fe composite preforms used in the metallurgical laboratory mainly depends on compacting load, aspect ratio, iron content, fractional density ratio and the die surface lubricant. Since these effects may not be linear and are usually interrelated, statistical methods are limited in their ability to predict the resulting process outcomes. Hence, the model was developed based on multi-layer Neural Network with a back propagation algorithm. Due to over-fitting, the conventional training method was not suitable to identify the required output parameters for unknown test data. Hence the standard tools like early stopping, regularization and Bayesian training were employed to enhance the neural network to recognize any independent test data.