KIET Journal of Computing and Information Sciences

Rice Varieties (LULC) Classification using Artificial Neural Network through Landsat 8 OLI Image

Talha Virk, MUHAMMAD USMAN — Sun, 12 Jan 2025 00:00:00 -0700

Multi-temporal and multi-spectral remote sensing images are good instruments that may help traditional agricultural systems by properly monitoring and calculating crop yields prior to harvesting. Traditional agricultural systems mostly depend on limited ground-survey data. Time-series satellite images of vegetation phenology serve a significant role in vegetation monitoring and land-cover categorization because they can record vegetation information at various development phases. It's impossible to overstate how important remote sensing has become in the last several decades. In order to monitor land surface dynamics, natural resource management, and the general status of the ecosystem, remote sensing sensors have been used from space. In order to begin vegetation conservation and restoration projects, it is necessary to know the existing status of plant cover. Several early studies of ecosystems and land cover have used agricultural census data to map paddy rice fields at the global and regional scales.

Social Distance Assessment and prevention system Based on Marker

Hira Beenish, Muhammad Fahad; Amta Nadeem, Syed Abbas Raza — Sun, 12 Jan 2025 00:00:00 -0700

IoT is quickly becoming a leading technology in healthcare. Early detection of health problems and preset protocols after patient recovery are all employed to decrease the chance of COVID-19 spreading to others in the event of COVID-19. Such wireless positioning devices can correctly remind individuals to maintain distance by sensing between people and then warning them if the people are close to each other. Motivated by this notion, in this paper we have proposed and implemented a model of the social distance assessment, monitoring, and marker system for prevention. The goal is to minimize the effects of the coronavirus outbreak while generating the least amount of economic harm possible, as well as to enable or even impose social distance. In the Monitoring System, users can easily access a web-based application that is integrated with the detection system by following the integration with the Raspberry Pi 4 and Pi Camera, in which they can monitor the detection of safe and unsafe people. Meanwhile, the marker system that is based on a laser will guide the user to stand in safer locations with the help of a laser marker module to eliminate violations. The proposed system is implemented using OPEN CV and Mobile NET SSD for object detection and uses the Euclidean Distance measurement method for measuring the distance between people. The hardware and software integration is also included in the system with an accuracy level, the system is an effective, low-cost, and user-friendly social spacing tool for preserving distance around people at a large gatherings.

Enhancing Histopathological Image Classification: Optimal Fine-Tuning of Convolutional Neural Networks with Feature Extraction

Sun, 12 Jan 2025 00:00:00 -0700

The area of medical image analysis has genuine obstacles, such as small data sets, complicated fine-tuning, and choosing the right architecture. This paper proposes a CancerVisionNet model based on the convolutional neural network (CNN) architecture with many layers to extract and classify features from cancer images. To train the CancerVisionNet model and avoid overfitting, data augmentation is carried out using a data set consisting of 220,025 images. The proposed CancerVisionNet model is evaluated on the PatchCamelyon dataset. Its remarkable area under the receiver operating characteristic (ROC) curve (AUC) measure is about 0.9. Compared to the other studies, the CancerVisionNet model stands out with a higher accuracy (95.4%). Moreover, this work demonstrates the potential of CNNs in medical image analysis, providing an effective approach to enhance classification accuracy and paving the way for further advances in the field. Although the results of this study pertain to histopathology and the PatchCamelyon dataset, the potential for a broader application awaits cross-domain validation. Future research works can include exploring alternative architectures and scalability to larger datasets.

A Prediction of Network Intrusion Using CNN-LSTM

Humza Rana, Farwa Zainab, Farwa Raoof, Attiya Zahoor — Sun, 12 Jan 2025 00:00:00 -0700

At present, network attacks have become a worldwide issue as they disturb the functioning and performance of the computer network. Network attacks are a serious problem they may cause the loss of important information, hacked personal data, and threats for demands money. Intruders today use more advanced type methods for hacking personal data and information as they break the traditional techniques used in the prevention of network intrusion. A more powerful and successful method is required for prediction. The deep learning method is suitable for this problem which is powerful and efficient in prediction. One of the famous types of deep learning models is the convolutional neural network model and another Long Short Term Memory. In this paper, the convolution neural network model combined with LSTM is proposed for the prediction of network intrusion which is one-dimensional. The proposed model is multiclass and tuned by different parameters to obtain the best efficiency from the model in the case of the multiclass dataset. This multiclass model is trained on the two multiclass datasets to get the best accuracy from the model on datasets. The first dataset named as wireless network dataset which contains four or five types of intrusion. The second dataset is the Microsoft Malware dataset that contains the eight or nine-class intrusion type. The experiment from the proposed model gives 0.990%, and 0.985% accuracy performance in multiclass prediction of network intrusion. The performance of the proposed hybrid CNN-LSTM model shows better performance than existing approaches

Topic Modeling and Identification in a Resource-Scarce Roman Urdu Language

Sun, 12 Jan 2025 00:00:00 -0700

In an age dominated by the Internet revolution, companies are making their businesses available to diverse groups of customers by leveraging the usage of e-commerce. To keep track of customer satisfaction and a competitive edge in the market, e-commerce businesses need to scrutinize their customer reviews. The manual approach to analyzing customer reviews is time and effort-consuming. Automated product review analysis exists but resource-poor languages like Roman Urdu lack such resources. To overcome this problem, this research presents a solution by incorporating Topic Modeling for Roman Urdu product reviews. A dataset of 8K Roman Urdu product reviews was curated from an online shopping platform. Various language-specific data cleaning steps were applied to data in the pre-processing step before experimentation. Different algorithms for Topic Modeling were implemented, out of them BERTopic produced outstanding results leaving the others behind. The results were also evaluated with an open-source dataset to check model generalization and reliability. Utilizing the power of machine learning and recent approaches, this study is a step forward to automated review analysis in the Roman Urdu language.