Ethiopian Banknote Recognition Using Convolutional Neural Network and Its Prototype Development Using Embedded Platform

Money transactions can be performed by automated self-service machines like ATMs for money deposits and withdrawals, banknote counters and coin counters, automatic vending machines, and automatic smart card charging machines. There are four important functions such as banknote recognition, counterfeit banknote detection, serial number recognition, and fitness classification which are furnished with these devices. Therefore, we need a robust system that can recognize banknotes and classify them into denominations that can be used in these automated machines. However, the most widely available banknote detectors are hardware systems that use optical and magnetic sensors to detect and validate banknotes. These banknote detectors are usually designed for specific country banknotes. Reprogramming such a system to detect banknotes is very difficult. In addition, researchers have developed banknote recognition systems using deep learning artificial intelligence technology like CNN and R-CNN. However, in these systems, dataset used for training is relatively small, and the accuracy of banknote recognition is found smaller. The existing systems also do not include implementation and its development using embedded systems. In this research work, we collected various Ethiopian currencies with different ages and conditions and applied various optimization techniques for CNN architects to identify the fake notes. Experimental analysis has been demonstrated with different models of CNN such as InceptionV3, MobileNetV2, XceptionNet, and ResNet50. MobileNetV2 with RMSProp optimization technique with batch size 32 is found to be a robust and reliable Ethiopian banknote detector and achieved superior accuracy of 96.4% in comparison to other CNN models. Selected model MobileNetV2 with RMSProp optimization has been implemented through an embedded platform by utilizing Raspberry Pi 3 B+ and other peripherals. Further, real-time identification of fake notes in a Web-based user interface (UI) has also been proposed in the research.

Deep Learning Model for the Automatic Classification of White Blood Cells

Blood cell count is highly useful in identifying the occurrence of a particular disease or ailment. To successfully measure the blood cell count, sophisticated equipment that makes use of invasive methods to acquire the blood cell slides or images is utilized. These blood cell images are subjected to various data analyzing techniques that count and classify the different types of blood cells. Nowadays, deep learning-based methods are in practice to analyze the data. These methods are less time-consuming and require less sophisticated equipment. This paper implements a deep learning (D.L) model that uses the DenseNet121 model to classify the different types of white blood cells (WBC). The DenseNet121 model is optimized with the preprocessing techniques of normalization and data augmentation. This model yielded an accuracy of 98.84%, a precision of 99.33%, a sensitivity of 98.85%, and a specificity of 99.61%. The proposed model is simulated with four batch sizes (BS) along with the Adam optimizer and 10 epochs. It is concluded from the results that the DenseNet121 model has outperformed with batch size 8 as compared to other batch sizes. The dataset has been taken from the Kaggle having 12,444 images with the images of 3120 eosinophils, 3103 lymphocytes, 3098 monocytes, and 3123 neutrophils. With such results, these models could be utilized for developing clinically useful solutions that are able to detect WBC in blood cell images.

Recognition of Gurmukhi Handwritten City Names Using Deep Learning and Cloud Computing

The challenges involved in the traditional cloud computing paradigms have prompted the development of architectures for the next generation cloud computing. The new cloud computing architectures can generate and handle huge amount of data, which was not possible to handle with the help of traditional architectures. Deep learning algorithms have the ability to process this huge amount of data and, thus, can now solve the problem of the next generation computing algorithms. Therefore, these days, deep learning has become the state-of-the-art approach for solving various tasks and most importantly in the field of recognition. In this work, recognition of city names is proposed. Recognition of handwritten city names is one of the potential research application areas in the field of postal automation For recognition using a segmentation-free approach (Holistic approach). This proposed work demystifies the role of convolutional neural network (CNN), which is one of the methods of deep learning technique. Proposed CNN model is trained, validated, and analyzed using Adam and stochastic gradient descent (SGD) optimizer with a batch size of 2, 4, and 8 and learning rate (LR) of 0.001, 0.01, and 0.1. The model is trained and validated on 10 different classes of the handwritten city names written in Gurmukhi script, where each class has 400 samples. Our analysis shows that the CNN model, using an Adam optimizer, batch size of 4, and a LR of 0.001, has achieved the best average validation accuracy of 99.13.

Breast Tumor Detection and Classification in Mammogram Images Using Modified YOLOv5 Network

Breast cancer incidence has been rising steadily during the past few decades. It is the second leading cause of death in women. If it is diagnosed early, there is a good possibility of recovery. Mammography is proven to be an excellent screening technique for breast tumor diagnosis, but its detection and classification in mammograms remain a significant challenge. Previous studies’ major limitation is an increase in false positive ratio (FPR) and false negative ratio (FNR), as well as a drop in Matthews correlation coefficient (MCC) value. A model that can lower FPR and FNR while increasing MCC value is required. To overcome prior research limitations, a modified network of YOLOv5 is used in this study to detect and classify breast tumors. Our research is conducted using publicly available datasets Curated Breast Imaging Subset of DDSM (CBIS-DDSM). The first step is to perform preprocessing, which includes image enhancing techniques and the removal of pectoral muscles and labels. The dataset is then annotated, augmented, and divided into 60% for training, 30% for validation, and 10% for testing. The experiment is then performed using a batch size of 8, a learning rate of 0.01, a momentum of 0.843, and an epoch value of 300. To evaluate the performance of our proposed model, our proposed model is compared with YOLOv3 and faster RCNN. The results show that our proposed model performs better than YOLOv3 and faster RCNN with 96% mAP, 93.50% MCC value, 96.50% accuracy, 0.04 FPR, and 0.03 FNR value. The results show that our suggested model successfully identifies and classifies breast tumors while also overcoming previous research limitations by lowering the FPR and FNR and boosting the MCC value.

MAINTAIN AGENT CONSISTENCY IN SURAKARTA CHESS USING DUELING DEEP NETWORK WITH INCREASING BATCH

Deep reinforcement learning usage in creating intelligent agents for various tasks has shown outstanding performance, particularly the Q-Learning algorithm. Deep Q-Network (DQN) is a reinforcement learning algorithm that combines the Q-Learning algorithm and deep neural networks as an approximator function. In the single-agent environment, the DQN model successfully surpasses human ability several times over. Still, when there are other agents in the environment, DQN may experience decreased performance. This research evaluated a DQN agent to play in the two-player traditional board game of Surakarta Chess. One of the drawbacks that we found when using DQN in two-player games is its consistency. The agent will experience performance degradation when facing different opponents. This research shows Dueling Deep Q-Network usage with increasing batch size can improve the agent's performance consistency. Our agent trained against a rule-based agent that acts based on the Surakarta Chess positional properties and was then evaluated using different rule-based agents. The best agent used Dueling DQN architecture with increasing batch size that produced a 57% average win rate against ten different agents after training for a short period. ABSTRAK: Pembelajaran Peneguhan Mendalam adalah terbaik apabila digunakan bagi mewujudkan ejen pintar dalam menyelesaikan pelbagai tugasan, terutama jika ia melibatkan algoritma Pembelajaran-Q. Algoritma Rangkaian-Q Mendalam (DQN) adalah Pembelajaran Peneguhan berasaskan gabungan algoritma Pembelajaran-Q dan rangkaian neural sebagai fungsi penghampiran. Melalui persekitaran ejen tunggal, model DQN telah beberapa kali berjaya mengatasi kemampuan manusia. Namun, ketika ejen lain berada dalam persekitaran ini, DQN mungkin kurang berjaya. Kajian ini melibatkan ejen DQN bermain papan tradisional iaitu Catur Surakarta dengan dua pemain. Salah satu kekurangan yang dijumpai adalah konsistensi. Ejen ini akan kurang bagus ketika berhadapan lawan berbeza. Kajian menunjukkan dengan penggunaan Rangkaian-Q Dwipertarungan Mendalam bersama peningkatan saiz kumpulan dapat meningkatkan konsistensi prestasi ejen. Ejen ini telah dilatih untuk melawan ejen lain berasaskan peraturan dan sifat kedudukan Catur Surakarta. Kemudian, ejen ini diuji berpandukan peraturan berbeza. Ejen terbaik adalah yang menggunakan rekaan DQN Dwipertarungan bersama peningkatan saiz kumpulan. Ianya berhasil memenangi permainan dengan purata 57% berbanding sepuluh agen lain melalui latihan jangka masa pendek.

Deep Convolutional Neural Network for Rice Density Prescription Map at Ripening Stage Using Unmanned Aerial Vehicle-Based Remotely Sensed Images

In this paper, UAV (unmanned aerial vehicle, DJI Phantom4RTK) and YOLOv4 (You Only Look Once) target detection deep neural network methods were employed to collected mature rice images and detect rice ears to produce a rice density prescription map. The YOLOv4 model was used for rice ear quick detection of rice images captured by a UAV. The Kriging interpolation algorithm was used in ArcGIS to make rice density prescription maps. Mature rice images collected by a UAV were marked manually and used to build the training and testing datasets. The resolution of the images was 300 × 300 pixels. The batch size was 2, and the initial learning rate was 0.01, and the mean average precision (mAP) of the best trained model was 98.84%. Exceptionally, the network ability to detect rice in different health states was also studied with a mAP of 95.42% in the no infection rice images set, 98.84% in the mild infection rice images set, 94.35% in the moderate infection rice images set, and 93.36% in the severe infection rice images set. According to the severity of rice sheath blight, which can cause rice leaves to wither and turn yellow, the blighted grain percentage increased and the thousand-grain weight decreased, the rice images were divided into these four infection levels. The ability of the network model (R2 = 0.844) was compared with traditional image processing segmentation methods (R2 = 0.396) based on color and morphology features and machine learning image segmentation method (Support Vector Machine, SVM R2 = 0.0817, and K-means R2 = 0.1949) for rice ear counting. The results highlight that the CNN has excellent robustness, and can generate a wide range of rice density prescription maps.

INVESTIGATION OF POINTNET FOR SEMANTIC SEGMENTATION OF LARGE-SCALE OUTDOOR POINT CLOUDS

Semantic segmentation of point clouds is indispensable for 3D scene understanding. Point clouds have credibility for capturing geometry of objects including shape, size, and orientation. Deep learning (DL) has been recognized as the most successful approach for image semantic segmentation. Applied to point clouds, performance of the many DL algorithms degrades, because point clouds are often sparse and have irregular data format. As a result, point clouds are regularly first transformed into voxel grids or image collections. PointNet was the first promising algorithm that feeds point clouds directly into the DL architecture. Although PointNet achieved remarkable performance on indoor point clouds, its performance has not been extensively studied in large-scale outdoor point clouds. So far, we know, no study on large-scale aerial point clouds investigates the sensitivity of the hyper-parameters used in the PointNet. This paper evaluates PointNet’s performance for semantic segmentation through three large-scale Airborne Laser Scanning (ALS) point clouds of urban environments. Reported results show that PointNet has potential in large-scale outdoor scene semantic segmentation. A remarkable limitation of PointNet is that it does not consider local structure induced by the metric space made by its local neighbors. Experiments exhibit PointNet is expressively sensitive to the hyper-parameters like batch-size, block partition and the number of points in a block. For an ALS dataset, we get significant difference between overall accuracies of 67.5% and 72.8%, for the block sizes of 5m × 5m and 10m × 10m, respectively. Results also discover that the performance of PointNet depends on the selection of input vectors.

DETERMINATION OF THE OPTIMAL BATCH, SIZE AND QUANTITY OF TRANSPORT PACKAGES FOR THE TRANSPORTATION OF KARELIAN BIRCH

This article presents the calculation of the optimal batch, dimensional and mass characteristics and the required number of transport packages for the transportation of Karelian birch in "knife" bars and boards. The number and weight of bars on EURO or FIN pallets, the number of boards of the same cubic capacity are determined and transport packages and stacks of boards are calculated taking into account the realization of the maximum possible carrying capacity and cargo capacity of vehicles and containers. The choice and justification of the rolling stock and its necessary quantity for transportation of the calculated batch of wood is made.

Towards a Theory of Quantum Gravity from Neural Networks

Neural network is a dynamical system described by two different types of degrees of freedom: fast-changing non-trainable variables (e.g., state of neurons) and slow-changing trainable variables (e.g., weights and biases). We show that the non-equilibrium dynamics of trainable variables can be described by the Madelung equations, if the number of neurons is fixed, and by the Schrodinger equation, if the learning system is capable of adjusting its own parameters such as the number of neurons, step size and mini-batch size. We argue that the Lorentz symmetries and curved space-time can emerge from the interplay between stochastic entropy production and entropy destruction due to learning. We show that the non-equilibrium dynamics of non-trainable variables can be described by the geodesic equation (in the emergent space-time) for localized states of neurons, and by the Einstein equations (with cosmological constant) for the entire network. We conclude that the quantum description of trainable variables and the gravitational description of non-trainable variables are dual in the sense that they provide alternative macroscopic descriptions of the same learning system, defined microscopically as a neural network.