To diagnose different ear diseases and disorders, otoscopy is an essential diagnostic method that looks at the eardrum and external ear canal. This project focuses on preprocessing, augmenting, and implementing Convolutional Neural Network (CNN) architectures for otoscopy image datasets, aiming to classify images such as normal, Acute Otitis Media (AOM), Tympanosclerosis, Acute Otitis Externa (AOE), Foreign Objects in the ear, and other conditions. The primary objective is to identify the architecture that demonstrates superior performance in terms of accuracy across different classes of ear conditions, achieved through fine-tuning and optimizing different layers. Among the various CNN architectures explored, the MobileNetV2 model exhibited notably high accuracy compared to others. Therefore, it was selected for deployment on a Raspberry Pi for real-world testing. A systematic approach to fine-tuning, focusing on optimizing key hyperparameters and architectural components and minor modifications to the base model architecture, the top layers of the base model are unfrozen to allow them to be fine-tuned on the disease identification task. Initially, the pre-trained MobileNetV2 model showed an accuracy of 66%. However, through fine-tuning and modification, the model's accuracy significantly improved to 97%, indicating the effectiveness of the proposed approach in enhancing classification performance. This study contributes to the advancement of automated otoscopy diagnosis by leveraging deep learning techniques, particularly CNN architectures. The successful deployment of the optimized MobileNetV2 model on a low-resource platform like Raspberry Pi underscores its potential for practical clinical applications, facilitating timely and accurate diagnosis of various ear conditions.

India is the largest contributor of cotton production in the world but at recent times the cotton production has been decreased significantly due to various cotton plant diseases. The bacterial and fungal diseases affect the growth of the cotton at early stages and cause decrease in cotton production. Manual monitoring of these plant diseases is not possible as cotton plants are cultivated in huge acres of land. Early detection of cotton plant diseases prevents the rapid spread of disease to the whole cotton field. This project deals with the implementation of a deep learning model to detect the type of cotton disease that affects the cotton plant and to classify whether it is a fresh cotton leaf or diseased cotton leaf at earlier stages of plant growth. MobileNetV2, a CNN based model is implemented with the real time dataset. When MobileNetV2 is compared to other CNN models, it performs better in terms of model size, accuracy, and validation speed, demonstrating its superiority in the classification and identification of diseases affecting cotton plants. Using drones for real time field monitoring is an efficient technique. It also reduces time consumption. In this project STM32 Discovery board and the camera module is integrated with the drone and the developed deep learning model has been deployed to it for real time classification of cotton plant diseases. The farmer can take appropriate preventive measures when plant diseases are detected at the earlier stage. Accuracy, loss, precision and recall of the model were analyzed and considered as the evaluation metrics of the developed model. Final output such as type of cotton disease affected and the accuracy of disease prediction is displayed to the user. The location of infected plant is tracked using the GPS module integrated with the microcontroller board.

Arrhythmogenic Right Ventricular Dysplasia is a heart disorder in which the heart muscle is replaced by fatty fibrous tissue. This causes sudden deaths in young men. The proposed novel AC-GAN and machine learning based ARVD detection system focuses mainly on detection of ARVD using ECG signals. This is the first machine learning based approach of detection of ARVD. This system involves creating synthetic ECG data of ARVD using AC-GAN. Since there is no available dataset in the internet, this approach results in large amount of synthetical ECG dataset. The primary characteristics of ARVD include inverted T waves and Epsilon waves at the end of the QRS complex. Various machine learning algorithm, such as the Decision Tree algorithm, Linear discriminant algorithm, Logistic regression algorithm, Naïve Bayes algorithm, RNN LSTM were trained and tested. The RNN LSTM has the highest tested accuracy of 89.2%. And was concluded to be the best fit for the ARVD detection system. The users can upload their ECG data to the web dashboard for real-time prediction.

The rapidly expanding Internet of Things (IoT) necessitates efficient and secure methods for firmware updates to improve device functionality, introduce new features and mitigate existing bugs. Firmware Over-the-Air (FOTA) is crucial in achieving these goals, employing diverse implementation techniques. However, evolving cybersecurity threats necessitate enhanced security measures in FOTA methods. This paper introduces a novel FOTA update approach integrating advanced encryption, compression techniques, and robust communication protocols to enhance the security and performance of the system. A comparative analysis with current state-of-the-art methods demonstrates the improved performance of the proposed solution. Key features include formal registration of IoT vendors and devices via a secure website, IPV6 VPN connectivity with TLS/SSL encryption, and IPFS-based retrieval of firmware updates using Content Identifiers (CIDs). Additionally, the paper discusses a conceptual approach for updating firmware in IoT devices without internet connectivity using Unmanned Aerial Vehicles (UAVs). Overall, the methodology addresses limitations of existing approaches and emphasizes strengthened security measures in FOTA operations. The proposed method not only enhances FOTA security but also ensures efficient and reliable firmware updates across diverse IoT ecosystems.

As numerous manufacturing enterprises are progressing towards Industry 4.0, advanced predictive models are required to optimize contemporary construction practices. The precise prediction of the compressive strength of cement is crucial, as it is an integral material for any constructional unit. This research paper explores numerous advanced machine learning and ensemble learning techniques for effective concrete strength prediction, enabling proactive quality control measures in an Industry 4.0 based environment. The research utilizes an open-source dataset and employs advanced machine learning techniques to interpret and learn intricate relationships among input features, such as cement quantity, blast furnace slag content, fly ash ratios, water weight, superplasticizer usage, and coarse and fine aggregate proportions, as well as curing age for predictive modeling. Experimental results validate the Histogram-Based Gradient Boosting model as an optimal technique for effectivey forecasting the compressive strength of cement in Newtons per square millimeter (MPa), with a cross-validation R2 Score of 0.922. The findings of this research work contributes to the increasing demand for accurate and scalable predictive models within the quality control unit of an Industry 4.0 based manufacturing firm.

In smart cities and urban environment, monitoring of water quality is very essential for environmental sustainability, resource optimization, cost management and streamlining the treatment process. In this research, an extensive dataset containing various contaminants such as aluminum, ammonia, arsenic, and others is utilized to effectively classify the safety level of water. Different machine learning and ensemble learning classifiers including LightGBM, XGBoost, CatBoost, Bagging, Gradient Boosting, Random Forest, Decision Tree, AdaBoost, MLP, and Extra Trees were implemented to conduct an empirical experimentation. Various performance metrics like Accuracy, Precision, Recall, F-1 Score, F-2 Score, Sensitivity, Specificity and AUC-ROC were used to evaluate the machine learning techniques utilized in this research. Experimental results indicate that the LightGBM ensemble technique outperformed other models with the accuracy of 97.13% and AUC-ROC of 98.99%, due to its optimized gradient boosting and effective processing of categorical features. This research contributes to the algorithmic approach for developing decision support tools for improving sustainable smart city water management techniques.

In this paper, we present a hybrid pareto optimality based lion swarm optimization and enhanced encryption standard technique for safe data transmission in WSN. Optimal, improved, and energy-efficient data transmission will be the emphasis of this endeavour to ensure proper tomato crop choices are made. In order to facilitate rapid and secure data transfer, this activity implemented multipath routing. In this study, we use a hybrid Pareto optimization technique based on the Lion Swarm Optimization (HPLSO) algorithm to quickly route data over many paths while satisfying a set of performance and latency requirements. Then, based on factors like propagation speed, distance, residual energy, and transmission range, the nodes grab the data packets for a time to prevent collisions and redundant packet transfers. The optimal node has a short retention period and is very likely to transmit data packets. If a node detects that data packets have been sent, it will simply delete them from the buffer. If the wait period hasn't expired, the packet shouldn't be sent on. The most efficient node in terms of power usage is chosen from among the nodes with the fewest packet losses using frontal Pareto optimization. In this paper, a new AES method using two keys is introduced, reducing the hacking difficulty while increasing the security of data transmission. Before being included to the forwarder's node list, each node in your region must register with the verification node, where its authentication will be validated. In terms of throughput, network lifetime, power consumption, packet loss rate, and latency, the results demonstrate that the proposed HPLSOAES architecture provides superior performance.

Road surface classification is very critical in the domain of intelligent driving systems as it ensures enhanced safety and comfort in autonomous vehicles. In this paper, a transfer learning approach is proposed for effectively classifying the road surface with enhanced accuracy and real-time performance mitigating the constraints of traditional image processing techniques A deep learning model is developed and trained in this research using an open-source dataset, encompassing 72,400 images capturing distinct road surface characteristics present across varying terrains. This research adopts pre-trained models like Mobilenet, VGG-19 and Resnet101 as feature extractors to optimize and improve the classification performance of the deep learning model significantly. The proposed methodology is evaluated using different key performance metrics such as loss, accuracy, average rate of improvement and change in accuracy, and total training time. The proposed deep learning technique achieved an accuracy of $\mathbf{9 8. 7 7 \%}$, signifying its critical importance for real-world applications and practical deployment.

A hybrid energy storage system (HESS) by integrating Lithium-Ion Battery and Wind Turbine System for Electric Vehicle is designed and implemented. An advanced model of lithium ion/wind turbine HESS model is developed to improve the battery’s charging capacity. The behavior of this model is tested by using Machine Learning Algorithms to identify remaining battery capacity of the energy storage components. The Machine learning Algorithms such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Random Forest (RF) Decision Tree (DT) and Linear Regression (LR) are implemented which helps in predicting the State of Charge (SoC) in the electric vehicle to estimate the battery’s lifetime. The SoC of the EV without HESS is calculated and compared with the SoC of HESS implemented EV model. Sensors are used to measure the current and voltage parameters of the Lithium-Ion Battery setup (Internal Battery) and Storage Battery (External Battery). The measured data is sent to the LPC2148 ARM microcontroller Board. The data that is obtained from LPC2148 is given as input to the machine learning algorithms which are executed in the Computer. Performance metrics such as Mean Absolute Error (MAE), Root Mean Square (RMSE), R-Square Error (R2E) and Mean Square Error (MSE) are calculated and the SoC values are predicted. The Random Forest algorithm produces better results for the HESS model as the Error values produced are less than 2.5%. The battery capacity of the electric vehicle has been improved by using external battery that is charged by wind turbine. Thus, the driving range of the electric vehicle implemented with HESS is increased twice when compared to conventional electric vehicle [1].

The agriculture is very essential and important part of human lives and serves as the basis of economical growth. The yield of the crops is dependent on numerous factors and the most important factor which effects the crop production in the soil. Soil has several parameters such as Nitrogen content(N), phosphorous content, potassium content(K), pH, temperature and humidity which are to be analysed in order to provide a view on which crop is suitable for cultivation. In the current research, a relative survey is made on the soil characteristic parameters in order to predict the future crop which can be grown is performed using predictive analysis algorithms. In addition to that the NPK content for each crop is analysed and if there is any deficiency in these contents, suggestions are given in order to improve the deficient content. The embedded system which has sensors and controller is used in the data acquisition process. The analysis is done using the data which are obtained from the sensors like Nitrogen, phosphorous, potassium, pH and temperature which are attached to the field and the sensor data is processed using the LPC2148 microcontroller. The data set are trained using KNN, Gaussian Naïve Bayes, Random Forest, Decision tree, Logistic regression and SVM. Experimental results indicate that Gaussian Naïve Bayes and Random Forest are found to be more efficient with reference to metrices like accuracy, precision, F1 score and recall. And hence the prediction is made using the GNB and Random Forest algorithm and the percentage of accuracy for the algorithms are 97.3

In this paper, a deep learning-based machine vision approach is proposed to automatically detect and classify defective tiles in the production assembly line of a tile manufacturing industry. The deep learning model used in this methodology is trained with 30,000 real-time images of cement/ceramic tiles, and the features of the image samples are extracted using the convolutional layers in the model. The defective tiles are identified and classified using an optimized activation function that acts as the decision-making layer or output layer of the deep learning model. The performance of this deep learning technique is evaluated using various metrics like accuracy, precision, recall and f1-score which is further compared with state-of-the-art activation functions like Relu, sigmoid, tanh and softmax. To further enhance the performance metrics, the feature extraction is done using various pre-trained models like VGG-16, Resnet50 and InceptionV3 and was further evaluated using metrics like K (Kappa statistic), OA (overall accuracy) and AA (average accuracy). The obtained experimental results with an accuracy of 99.96% under a favorable learning rate prove the robustness and efficiency of the proposed methodology to enhance industrial quality control in any tile manufacturing industry.

The rapid expansion of autonomous driving technologies necessitates the development of robust systems for accurate road surface identification and classification to ensure safe and reliable driving. This review article addresses the imperative requirement for effective road surface classification by investigating diverse methodologies within the realms of computer vision and the Internet of Things (IoT). Through an extensive investigation, various techniques encompassing Image processing, Machine Learning(ML), Deep Learning(DL), and IoT are examined for their effectiveness in classifying road surfaces at different terrains. Moreover, this article also reviews datasets, signals, sensors, communication protocols, IoT implementation strategies, pre-processing methodologies, and feature extraction techniques. This investigation delves into novel approaches aimed at resolving road surface classification challenges, meticulously examining their respective strengths and limitations. Furthermore, this article includes a comparative analysis of these advanced methodologies, facilitating the identification of the most suitable model for the task. The assessment takes into account intricate methodological aspects, types of sensors/cameras, dataset variations, and performance metrics, thereby providing valuable insights into the landscape of road surface classification for autonomous driving applications.

This research study proposes a medical gas pressure monitoring system that is developed to monitor the gas pressure level of each gas channel by measuring their pressure using an ABP series mount Pressure sensor. One of the greatest provocations happening in the health sector is the absence of a proper monitoring system of gas supply for patients. The proposed system uses a PIC16F877A microcontroller as the controller board. Totally three gas channels are taken into consideration. The pressure value of each gas channel will be displayed based on the unit which will be set by the user. The units are PSI, Kg, and bar by default it will display in PSI unit and threshold pressure levels of below 50 as a low level and above 75 as a high level, for each pressure level of low and high respected led will blink as an indication. Whenever the gas pressure level is reduced below 50 the led and buzzer will be on, the proposed system has a mute button and mode button, and the user can mute it down using the mute button, and goes like this for pressure levels above 75. If the value stays at the pressure level of below 50 or above 75 for more than 15 minutes again buzzer will be on. And by using the mode button the user can vary the threshold of all three gas pressure levels using the mode buttons provided in the controller unit. This pressure level of a single channel at the particular voltage and current level is acquired as a raw real-time dataset without any feature extraction. This raw data which contains real-time parameters such as pressure, voltage, and current is acquired at the microcontroller and as well at the gas outlet pipe from a single channel (oxygen gas cylinder). This study uses ML algorithms such as KNN, logistic regression, random forest, naïve Bayes, and decision trees that get trained by the raw data and will give the parameter matrices such as precision, accuracy, F1 score, and recall to evaluate the algorithm. Hence, the decision tree algorithm achieved the highest performance rate of 81%. So, the decision tree algorithm is found to be the best Machine learning model which got well trained by the raw train dataset.

The proposed approach in this paper utilizes a combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) cell network based deep learning algorithm to forecast the optimal time period for replacing aircraft tyres. The solution compares various optimizers such as Nadam and Adam, and activation functions such as ReLU and Swish, in order to select the best suited model. The primary concern is accuracy, which is evaluated using the Root Mean Squared Error (RMSE) metric. The proposed CNN-LSTM algorithm can predict the required features with a RMSE of 0.3. The model that incorporates the NADAM optimizer and the Swish activation function exhibits a comparatively lower RMSE curve in the graph.

Transfer of data over the blockchain is significantly improved over the years. This proposed system gives a unique method of transferring patient health data to another hospital or to a peer, using a blockchain-based web application that places patient ownership and data security at the forefront. Patient data is stored in a decentralized and distributed manner using the Interplanetary File System (IPFS), eliminating the need for a central authority. Smart contracts written in Solidity facilitate transactions by checking for sufficient funds and managing the transaction process. A beacon proxy is used to automatically update patient ownership in the smart contract, reducing errors and inaccuracies resulting from manual intervention. The blockchain network offers a secure and decentralized solution for transferring patient health data, ensuring privacy and accessibility only to authorized individuals through advanced cryptographic techniques, distributed data storage, and smart contracts.

Due to the lack of efficient transportation and infrastructure, 16

Tea is one of the popular and widely cultivated plantation crops in Tamil Nadu. Their productions are affected by different diseases that affect them. The quality of the tea leaves and bud needs to be monitored to increase the profit. In this paper a model is implemented using deep learning and machine learning techniques to detect the common diseases that affect the tea leaves and to classify the tea bud whether they are in the right stage to be plucked or not which signifies the high-quality and low-quality bud respectively. In this project two algorithms namely CNN and SVM are implemented using real time dataset. Since deep learning requires a large dataset for training. 1050 images for tea leaves disease detection and 262 images for tea bud classification are collected in real time and augmentation on these images is carried out to increase the number of input images. In CNN Dense-Net Architecture 121 and 201 are implemented for disease and bud classification which with deep convolution layers gives a high accuracy of 96.703% and 96.923% respectively. In SVM the same convolution model is built and converted into an SVM kernel which gives better results than regular implementation with an accuracy as 65.934% and 73.077% for disease and bud classification respectively. Evaluation metrics such as Accuracy, Loss, Precision, Recall, F1 Score, Mean Square Error, Mean Absolute Error, Specificity and Sensitivity of the model were analyzed

The paper proposes energy prediction techniques on Electric energy consumption from existing electricity supply using machine learning. Machine learning and its techniques are vastly used in factories for maintenance, in industries for optimization and to calculate the fault tolerance to produce an efficient monitoring system. This paper proposes the best model to monitor the usage of power for predicting the energy Raspberry Pi 3Model B+ is the core controller used for this energy prediction and modelling. AC-current, AC-voltage and power are the input parameters measured using current sensor (ACS712) and AC-Voltage sensor (ZMPT101B). MCP3008 ADC is interfaced with the controller to read analogue current and voltage values. Linear regression and logistic regression are used for both prediction and optimization. The two algorithms are implemented on the Spyder IDE which is installed in the Raspberry pi 3Model B+ controller. After reading the input parameter it predicts the energy consumption and displays whether it is high or low for that day. Experimental Results suggests that the linear regression model is the best among the two because the accuracy metric values viz., Mean Absolute Error, R2 score, Execution time, Residual Sum of Square are comparatively less when compared with logistic regression model.

Stress is a universal emotion that every human experiences daily. Psychologists say stress may lead to heart attack, depression, hypertension, strokes, or even sudden death. Many technical explorations like stress detection through facial expression, speech, text, physical behaviors, etc., were explored, but no consensus has been reached on the best method. The advancement in biomedical engineering yielded a rapid development of electroencephalogram (EEG) signal analysis that has inspired the idea of a multimethod fusion approach for the first time which employs multiple techniques such as discrete wavelet transform (DWT) for de-noising, adaptive synthetic sampling (ADASYN) for class balancing, and affinity propagation (AP) as a stratified sampling model along with the artificial neural network (ANN) as the classifier model for human emotion classification. From the EEG recordings of the DEAP dataset, the artifacts are removed, the signal is decomposed using a DWT, and features are extracted and fused to form the feature vector. As the dataset is high-dimensional, feature selection is done and ADASYN is used to address the imbalance of classes resulting in large-scale data. The innovative idea of the proposed system is to perform sampling using affinity propagation as a stratified sampling-based clustering algorithm as it determines the number of representative samples automatically which makes it superior to the K-Means, K-Medoid, that requires the K-value. Those samples are used as inputs to various classification models, the comparison of the AP-ANN, AP-SVM, and AP-RF is done, and their most important five performance metrics such as accuracy, precision, recall, F1-score, and specificity were compared. From our experiment, the AP-ANN model provides better accuracy of 86.8% and greater precision of 85.7%, a higher F1 score of 84.9%, a recall rate of 84.1%, and a specificity value of 89.2% which altogether provides better results than the other existing algorithms.