Over the years significant research has been performed for machine vision based fabric inspection systems in order to replace manual inspection, which is time consuming and not accurate enough. Automated fabric inspection systems mainly involve two challenging problems: one is defect detection and another is classification, which remains elusive despite considerable research effort in automated fabric inspection. The research reported to date to solve the defect classification problem appears to be insufficient, particularly in selecting appropriate set of features. Scene analysis and feature selection play a very important role in the classification process. Insufficient scene analysis results in an inappropriate set of features. Selection of an inappropriate feature set increases complexities of subsequent steps and makes the classification task harder. Considering this observation, we present a possibly appropriate feature set in order to address the problem of fabric defect classification using neural network (NN). We justify the features from the point of view of distinguishing quality and feature extraction difficulty. We performed some experiments in order to show the utility of proposed features and compare performances with recently reported relevant works. More than 98% classification accuracy has been found, which appears to be very promising

Zero-Day attack detection in Network Intrusion Detection Systems (NIDS) refers to the ability to identify previously unseen attack patterns during testing without having been explicitly trained on those specific attacks, utilizing learned features from other known attacks. In this paper, we propose a Deep Reinforcement Learning (DRL)-based NIDS designed for Zero-Day attack detection. We use a stacked LSTM architecture to extend the learning capabilities of the DRL agent. We apply several oversampling techniques to handle the issue of class imbalance since the zero-day attack datasets are not as abundant. We use some of the most widely available benchmark datasets in NIDS domain, which all together cover a wide range of attack types, such as reconnaissance, ddoS, infiltration, injection, password attacks, brute force, dos, backdoor, and benign traffic. For example, we converted attacks to 1 and benign traffic to 0, then excluded certain attack categories (DoS and Backdoor) from the training dataset while keeping them in the test dataset. This makes those attack types zero-day attacks, as they are entirely unseen during training. We also compare which data balancing technique works better among K-means SMOTE, SMOTE, Borderline-SMOTE and ADASYN on the performance of our DRL agent. We then demonstrate how powerful our agent is by validating many datasets for remarkable success in detecting both known and unknown attacks in a zero-day manner.

Bangladesh is an agricultural country having a tropical monsoon climate. A large variety of tropical and sub-tropical fruits abound in Bangladesh. People of Bangladesh are fruit-lovers too. Currently, most of the people of this country are failing to recognize many of the rare local fruits and the number of this portion of people is increasing day by day. Thus, not only the natural heritage but also good sources of food are being diminished. Performing a machine vision based recognition of these fruits can help people recognize them. In this paper, we perform an in-depth exploration of a computer vision approach for recognizing rare local fruits of Bangladesh. A number of rare local fruits are classified based on the features extracted from their images. For our experiment, we have used a total of 480 images of 6 rare local fruits. We perform some preprocessing on the captured image and then expected features are extracted using image segmentation. Classification of the fruits is accomplished using support vector machines (SVMs). We have achieved 94.79% classification accuracy, which is not only good but also promising for future research.

Textile inspection system, which carries a lot of importance in the production process of textile goods, has been part of a great deal of research for automating the process. Manual textile inspection is a lengthy, slow as well as erroneous job; therefore, automation of textile inspection is a demand of time. Machine vision based i.e. automated fabric inspection deals with two primary challenges, namely defect detection and defect classification. The quality and quantity of research done on defect detection and classification is still not up to the mark. Our focus is on detecting various defects in textile fabrics and classifying them. We extracted features using statistical techniques. Images of textile fabrics were used as sample. Inspecting the images, geometric and statistical features were found out. Through this paper, we bring in a suitable Bayesian classifier to classify the images into different classes of defective properties. Our approach has delivered acceptable accuracy compared to other works in the domain of textile defect detection and classification.

Potatoes are one of the world's most popular and economically important crops. For many uses in agriculture, breeding, and trading, accurate recognition of potato breeds is important. In recent years, deep learning algorithms have become effective tools for breed recognition tasks using pictures, which inspires researchers to explore their potential for recognizing potato breeds. The paper presents extensive research on the application of deep learning for potato breed recognition. The recognition of potatoes has been effectively performed using the five state-of-the-art deep learning models VGG16, ResNet50, Mobile-Net, Inception-v3, and a customized CNN. These models have been modeled to differentiate between several potato breeds based on their unique visual characteristics, such as size, shape, color, texture, and skin pattern, by being trained on images of various potato breeds. The performance of each of the deep learning models is evaluated through thorough evaluation and testing. Among the models, the customized CNN model gives the best accuracy. The customized CNN model's accuracy is 94.84%. We do not just evaluate the accuracy but rather some other indicative metrics, such as F1-score, recall, and precision, too.