Adequate air ventilation systems and maintaining environmental air quality are essential things that need to be considered to create an excellent industrial environment in a company. Operating or maintaining ecological air quality is a challenge for companies to optimize the existing work environment. Especially at the economic and business level in the company are facing the main problems. In this case, monitoring and predicting future air quality are needed to maintain air quality conditions. PM10 concentration, humidity, and temperature were used to predict annual and seasonal indoor air quality using a recurrent neural network (RNN) and long short‐term memory (LSTM). In this paper, we propose the IAQP method for air quality management systems that combines indoor air quality forecasting based on real‐time data. To measure indoor air quality, we predict the outcome from the IIoT sensor and LoRa sensor. The result of prediction is that the multilevel RNN model outperformed the LSTM, as the model demonstrated excellent results and feasibility.

This technical paper proposes a new waveform combining the low-rate and high-rate data streams to detect the region-of-interest signal in a high-mobility environment using optical camera communication. The proposed technique augments the bit rate of the low-rate stream; consequently, the link setup time is reduced and the requirement of low frame rate camera is eliminated. Additionally, both the low-rate and high-rate data streams in the proposed bi-level pulse position modulation are decoded with a unique adaptive thresholding mechanism with a high frame rate camera. We also propose a vehicle localization scheme to assist the drivers in maintaining a safe following distance that can significantly reduce the frequency of accidents. Moreover, two neural networks are proposed to detect the light-emitting diodes (LEDs) for localization and communication, and to estimate the road curvature from different rear LED shapes of the forwarding vehicle, respectively. The system is implemented, and its performance is analyzed in Python 3.7. The implementation results show that the proposed system is able to achieve 75% localization accuracy, a 150 bps low-rate stream, and a 600 bps high-rate stream over a range of 25 m with a commercial 30 fps camera.

Automated betel leaf disease recognition remains challenging under limited and imbalanced data conditions. This study investigates an Upper Confidence Bound (UCB)-guided adaptive augmentation framework for the classification of healthy leaves, leaf rot, and leaf spot. To reduce the risk of performance inflation caused by augmented-data leakage, only the 2037 original images of a publicly available betel leaf disease dataset were used for stratified training, validation, and test partitioning, while augmentation was restricted to the training stage. Several representative deep learning architectures, including ResNet, DenseNet, EfficientNet, MobileNet, and ConvNeXt-Tiny, were evaluated under a unified experimental setting. In the proposed approach, augmentation-policy selection is formulated as a multi-armed bandit problem, where the UCB controller dynamically selects among predefined augmentation arms comprising different RandAugment magnitudes and Mixup settings based on validation performance. Among the evaluated configurations, ConvNeXt-Tiny integrated with UCB-guided augmentation achieved strong overall performance, with an accuracy of 98.04%, a macro-F1 score of 97.87%, and a micro-average AUC of 0.999. These findings indicate that combining ConvNeXt-Tiny with adaptive augmentation can provide an effective approach for betel leaf disease classification under the investigated limited-data setting, with potential for future deployment in resource-constrained agricultural monitoring applications.

In the area of image processing and computer vision, visual anomaly detection is a critical and difficult task. For anomaly detection in surface image data, a customized neural network incorporating self-gated rectified linear unit (SGReLU) was designed, and the SGReLU-based model excelled other activation function-based models with a top-20 average test accuracy of 99.84%. The computational time needed for the operation is 10533 s for 20 epochs and the top-20 average test loss is 0.0125 using SGReLU, both of them were comparatively less than other activation functions.