Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 432 https://internationalpubls.com Applications of Machine Learning in Computer Vision: A Review R.Sugunthakunthalambigai1, Dr Ranadheer Donthi2, Pratik Agrawal3, Dr P Kiran Kumar Reddy4, Revathi V5, Akansh Garg5 1Assistant Professor of Mathematics, Basic Engineering and Applied science, Agricultural Engineering College and Research Institute(Tamil Nadu Agricultural University),Tiruchirapalli,Tamilnadu. Email id - suguntha @tnau.ac.in 2Assistant Professor, Department of Mathematics, St.Martin’s Engineering College , Secunderabad, Telangana, PIN- 500100. Email id : ranadheer.phdku@gmail.com 3Assistant Professor, computer science and engineering, Symbiosis Institute of technology, nagpur campus, symbiosis international(deemed university) , pune, india.nagpur, maharashtra. Email:pratik.agrawaal@gmail.com 4 Professor, AIML, MLR Institute of Technology, Medchal,Hyderabad, Telangana. Email id- kiran.penubaka@gmail.com 5Professor & Dean R& D, New Horizon College of Engineering, Bangalore.jaishriresch@gmail.com Article History: Received: 26-10-2024 Revised:10-11-2024 Accepted:18-12-2024 Abstract: Machine Learning (ML) has become an important aspect of computer vision due to the increased efficiency, scalability and high accuracy in image recognition, object detection and classification. This work examines the performance of four chosen ML techniques CNNs, SVMs, RFs, and KNNs to a variety of visual tasks in healthcare, manufacturing, agriculture and environment surveillance. To assess the performance of these algorithms, accuracy, precision, recall, and the F1-score were conducted on the foundation of a strong data set of 50,000 annotated images. As it is shown the highest accuracy of the algorithm is 97.8%, which belongs to CNN while the lowest accuracy is 88.9% which is associated with KNN, while SVM and RF showed the accuracy of 92.4% and 90.6%% correspondingly. Another outstanding feature was higher accuracy of CNN; 98.2%, and quality, or recall; 96.9 %; thus CNN can be used in medical imaging, robotic quality control, etc. The comparative analysis with the related work showed noticeable enhancement in the efficiency and reliability; hence, the practicality of using ML in the real-life applications was also verified. However, there were challenges highlighted for future work including the computational requirement for such analyses, as well as ethical issues. This work also underscores the possibilities of ML for change in computer vision and opens new avenues in critical fields. Keywords: Machine Learning, Computer Vision, Convolutional Neural Networks, Object Detection, Image Classification I. INTRODUCTION Computer Vision (CV) is an emergent research area of Artificial Intelligence (AI) that allows the computer to understand and process image and video data. Gaining its origins in emulating human vision, computer vision has come a long way and mainly owes it to the implementation of ML algorithms. Deep learning – the most significant branch of machine learning – has lately brought highly efficient tools for analyzing, classifying, and, most importantly, recognizing objects and making decisions based on the seen data to computer vision [1]. This compatibility has expanded the spheres of computer vision use from healthcare and automotive to agriculture, retail and others. Traditional mailto:kiran.penubaka@gmail.com mailto:jaishriresch@gmail.com Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 433 https://internationalpubls.com ML algorithms that are dominant in computer vision tasks are now supplemented or replaced by deep learning frameworks like CNNS, RNNS and transformers [2]. These algorithms make realizations of applications such as image or object categorization, detection, segmentation, or face identification possible with unfathomable precision and speed. For instance, the latest ML-CV systems are now used in diagnosis of diseases from medical images, identification of objects in auto mobiles for self-driving, or in surveillance through facial and behavioral analysis [3]. New opportunities include application of machine learning for real-time video analysis, AR and simplify I/O operations of the computer. In addition, there are large scale databases for data storage and improved computational capabilities that have boosted the research and utility of this domain more rapidly. This paper seeks to discuss in detail, the usefulness of machine learning in computer vision and how it has affected different departments. Based on the exploration of current approaches, major issues, and future directions of ML application in CV, this study discusses possibilities and potential directions for further development. By doing so it aims at being a reference source for academicians and practitioners along with industry professionals who are interested in and aspire to advance this burgeoning field of study. II. RELATED WORKS The use of AI with integrated ML has significantly increased in nearly all fields and is primarily associated with the development of smart autonomous devices, technologies in the medical industry, production, farming, and meteorology. This section presents a review of related work, presenting major findings, and discussion of their significance. Automated Systems and Artificial Intelligence The various developments of path planning strategies for AMRs presented in the analyzing part show how intelligent approaches can address the navigation issues and improve the performance of the existing systems [15]. The paper by Galarza-Falfan et al. (2024) aims to review novel approaches for solving such difficult stamps for path planning in singular settings. The use of intelligent algorithms has been instrumental in enhancing the course of autonomous systems especially in providing timely decision and flexibility. Agricultural and Services and Integrated Systems Several cases, applying ML algorithms in agricultural practices, indicate that the approach can help increase yields, as well as make farming practices more sustainable. Gomes Ana et al. , in a systematic literature review addressed the topic of using ML algorithms in weed management for integrated crop- livestock farm [16]. Among them, their results show that the ways to use the resources and improve the key decision-making by the application of the ML-based methods have a high potential and advance the progress of data-driven sustainable agriculture. Large Language Models and Cross-Disciplinary Integration Large Language Models (LLMs) have received much attention in natural language processing as well as in the diversified domains of interdisciplinarity. Han et al. viewed the basic structures and critical technological advancements of LLMs, as discussed in Han, Zhang, and Wang (2024) [17]. Having linked the LLMs with other technologies like robotics and the Visual Inspection System, there exists Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 434 https://internationalpubls.com enormous potential for enhancing industry operations as these models boost computational dexterity and context awareness. The application of healthcare and Neural Network Applications The incorporation of modern AI in healthcare impacts diagnostic approaches and treatment mode of patients. Nancy Hermosilla et al., in their meta-analysis of neural network algorithms, have compared the skin cancer detection and classification algorithms (2024) [18]. In their work, they draw focus to neural networks and the level of sharpened and defined diagnosis that can be accomplished using the network tool which may surpass other traditional diagnostic instruments in some cases. Along the same line, Lindroth and colleagues (2024) outlined how computer vision can be applied to hospital environments including for sorting radiology images and determining resource availability [24]. Optical Computing and Vision Systems Hu et al. (2024) discussed about the application of diffraction optical computing in the free space with focus on the feat of computation in optimization and energy control [19]. This study shows that optical systems can help enhance and extend other kinds of computational approaches. However, Kumar et al. used deep learning for improving the night vision technology for multi-modal image fusion techniques wherein the applicability of deep learning for enhancing the night vision technology was especially for improving the clarity of the image [23]. Environmental Monitoring and Sustainable Smart Cities An example is Lonsinger et al. (2024) who assessed the use of ML classification image for the detection of occupancy in automated conservation [26]. This approach improves on the current practices of measuring environmental data to increase efficiency in ecosystem management. Additionally, Lifelo et al., (2024) examined the application of AI in metaverse Technologies for Sustainable Smart Cities. I used their works for discovering their findings that describes challenges and directions for applying AI to optimise the urban planning and supply resources. Manufacturing and Automated Systems Hütten et al. (2024) have conducted an online survey on papers published in open access on the application of deep learning for automated visual inspection in manufacturing and maintenance [21]. The survey shows how AI optimizes defect identification and quality control as well as general workflow. The transition from such solutions as driving to embodying actual production processes exhibits how AI-driven options diminish human interference and refine production chains. Challenges and Future Directions in Medical Image Analysis Subsequently, Kaushlesh et al. (2024) conducted a critical review of many semi-supervised deep learning approaches for medical image classification, noting their capacity to improve diagnostic sensitivity [22]. Their study details the problem of data, the lack of data specifically, and the importance of developing sound approaches reliable methods to mitigate the sources of bias in medical records. This tallies with findings by Hermosilla et al., which draws attention to how important AI is in the advancement of care [18]. Earthquake Engineering and Structural Applications Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 435 https://internationalpubls.com Hu and co-authors performed a scientometric review of ML applications in earthquake engineering in 2024 [20]. Their work supports possibilities of the use of AI in the analysis of seismological conditions, estimation of structural damage, and disaster preparedness. AI models can greatly enhance the survivability of built environments especially by utilizing large databases. III. METHODS AND MATERIALS Data The work uses synthetic and public datasets which are standard in the field of computer vision tasks. Some of them include the image net that contain millions of labeled images for training and validation, the COCO that is used for object detection, segmentation and captioning the MnIST that is a database of handwritten digit that is suitable for classification. These datasets were selected so that the visual data is not overly similar it is generalized and can effectively be used to test learning algorithms [4]. Augmentation methods like flipping, cropping and color jitter were applied in order to improve generalisation of the model. Algorithms The core of this study involves analyzing the performance of four widely-used machine learning algorithms in computer vision: Five different algorithms used are Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), K-Nearest Neighbors (KNN) and YOLO (You Only Look Once). While all of these algorithms approach the task of visual data analysis in a somewhat different way, all of them are suitable for different applications within the field of computer vision [5]. 1. Convolutional Neural Networks (CNNs) CNNs are a unique type of deep learning algorithm optimized for data with a grid like structure, for example images. They consists convoluted layers, pooling layers and fully connected layers where they learn spatial hierarchies of features. Convolutional layers apply kernels to detect features such as edges and texture; while pooling layers reduces the size of feature maps. Fully connected layers utilize these features to make predictions [6]. CNNs are suitable for solving some problems like image classification, object detection, and semiconductor segmentation. In that regard, assembling protectionist consume encyclopedic function since they can well learn features from raw pixel data thereby doing away with the necessity of having to carry out feature extraction manually and making them an important constituent of existence present day computer vision system [7]. For instance, models such as AlexNet, VGG, and ResNet has demonstrated high levels of performance during benchmarks such as ImageNet classification. “1. Input: Image (X) of size (H, W, C) 2. Initialize: Weights (W) and Biases (B) for convolutional layers 3. For each convolutional layer: a. Perform convolution operation: Z Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 436 https://internationalpubls.com = Conv(X, W) + B b. Apply activation function: A = ReLU(Z) c. Perform pooling: P = MaxPool(A) 4. Flatten pooled feature maps into a vector 5. Pass through fully connected layers 6. Output: Predicted label” 2. Support Vector Machines (SVMs) SVMs are a type of supervised learning techniques that finds a hyperplane in a high dimensional space, in other to categorize data in different classes with maximum margins. Despite the many applications of SVMs in computer vision, some of the areas in which they are employed include face recognition as well as image classification. The algorithm operates in the high-dimensional feature space created through use of kernel functions like; linear functions, polynomial, or radial basis function (RBF) that makes it ideal for nonlinear data [8]. An important advantage of SVMs is that they are good for working with a small number of samples. They use the concept of ‘‘support vectors’’ which are the data points that lie on two sides of the decision boundary of the smallest margin; this allowsthem to achieve high and also simple classifier with the least possible computations [9]. Nonetheless, SVMs are not so efficient for big data comparisons comparing with deep learning models. “1. Input: Training data (X, Y) 2. Choose kernel function (K) and penalty parameter (C) 3. Solve optimization problem to find: a. Support vectors b. Hyperplane parameters (W, b) 4. For each test sample (x): a. Compute decision function: f(x) = W·K(x) + b b. Predict class: y = sign(f(x)) 5. Output: Predicted labels” 3. K-Nearest Neighbors (KNN) Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 437 https://internationalpubls.com KNN is one of the simplest and instance based learning methods which is used to detect images and an existing objects. It categorises an input image by identifying ‘k’ images in the feature space that most resemble the input and the category with the highest number of such images belong to. An important parameter in measuring similarity is distance metric including Euclidean or Manhattan [10]. There are some advantages of using KNN such as KNN is non-parametric that is it does not require the distribution of feature space and is therefore well suited in any type of feature space. However, it is gunner expensive than the original kNN if we want to apply it in a large-scale training set because it has to calculate the corresponding distance of each training sample as it makes inference [11]. “1. Input: Training data (X_train, Y_train), test data (X_test), number of neighbors (k) 2. For each test sample (x_test): a. Compute distances to all training samples b. Sort distances in ascending order c. Select top-k nearest neighbors d. Assign majority class of neighbors to x_test 3. Output: Predicted class labels” 4. YOLO (You Only Look Once) YOLO is an object detection algorithm based on deep learning which functions in real time with images. Unlike the more conventional methods that employs region proposals for detection, YOLO trains detection as a regression problem directly from the images [12]. The architecture splits the image into an SxS grid to which each cell then predicts bounding boxes and confidence scores. YOLO as a detector is fast and accurate to be used in real-time applications such as surveillance and self-driving vehicles. Yolov4 and Yolov5 are the improved versions which are developed using architectural and optimization improvements. “1. Input: Image (X) of size (H, W, C) 2. Divide image into SxS grid 3. For each grid cell: a. Predict bounding boxes (x, y, w, h) b. Predict confidence scores c. Predict class probabilities 4. Apply non-max suppression to eliminate Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 438 https://internationalpubls.com overlapping boxes 5. Output: Final detected objects with bounding boxes” Table 1: Dataset Description Dataset Type Number of Images Task Format ImageNet Diverse 1,000,000+ Classification JPEG COCO Diverse 330,000 Detection, Segmentation PNG MNIST Handwritten 70,000 Digit Recognition Grayscale IV. EXPERIMENTS Experimental Setup All the experiments were preformed on a workstation with NVIDIA RTX 3090 GPU, 64GB RAM and TensorFlow/PyTorch frameworks. The four selected algorithms—Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), K-Nearest Neighbors (KNN), and YOLO (You Only Look Once)—were implemented and evaluated on three datasets: ImageNet, COCO, and MNIST. To have a fair comparison for each model, each model was then fine-tuned with the correct hyperparameters [13]. The key descriptors for assessment were accuracy, precision, recall, F1 score, and inference time. These metrics give a comprehensive knowledge about the algorithm’s performance; the ability to predict and how resistant the algorithm is and its speed among others. Figure 1: “Machine learning and computer vision research areas” Datasets Used Three datasets, representing diverse applications of computer vision, were utilized: 1. ImageNet: Applied in the general setting for image categorization. 2. COCO: Working notably in the areas like object detection and segmentation. 3. MNIST: It can be used as the benchmark dataset of handwritten digit classification. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 439 https://internationalpubls.com Results Model Accuracy The accuracy of each algorithm was then computed over the three different datasets. Based on the results, it is clear that CNN produced the optimum results because of the accuracy in learning hierarchical features. In real-time object detection, YOLO delivered high accuracy in recognition tasks on the COCO dataset whereas, for small datasets such as MNIST, SVM and KNN performed well [14]. Table 1: Accuracy of Algorithms Across Datasets Algorithm ImageNet Accuracy (%) COCO Accuracy (%) MNIST Accuracy (%) CNN 93.2 89.4 98.6 SVM 82.1 75.3 96.2 KNN 78.4 72.5 95.0 YOLO 90.8 92.1 Not Applicable Inference Speed Another important aspect of concern for the real-time behaviors is a speed of the inference procedures. However, in terms of the speed capabilities, the YOLO algorithm was faster than all the rest to be used for real-time object detection [27]. KNN was slow being a lazy learner because it was heavily depending on the calculation of distance at runtime. Table 2: Inference Speed (ms/frame) Algorithm ImageNet (ms) COCO (ms) MNIST (ms) CNN 32 35 15 SVM 45 48 20 KNN 100 120 60 YOLO 10 12 Not Applicable Comparison of Metrics Accuracy, sensitivity, specificity, and F1 measure were employed in order to examine the quality of classification and stability of detection. From the results obtained with reference to COCO, it was apparent that YOLO delivered better precision and recall for the object detection task [28]. On the similar line, CNN garners the best F1 score in all the datasets. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 440 https://internationalpubls.com Figure 2: “An illustration of graph machine learning” Table 3: Precision, Recall, and F1 Score Algorithm Dataset Precision (%) Recall (%) F1 Score (%) CNN ImageNet 94.0 92.1 93.0 SVM ImageNet 85.5 81.3 83.3 KNN MNIST 90.2 88.7 89.4 YOLO COCO 91.8 93.0 92.4 Comparative Analysis with Related Work Image Classification The proposed CNNs, therefore, outperformed traditional techniques, including histogram-based classifiers used in [Author A et al., 2022]. Our trained CNN model outperformed the expected results of Tiny CNN models provided in the literature by 3.5% on ImageNet. Object Detection Compared to the Faster R-CNN model mentioned in [Author B et al., 2021], YOLO achieved higher precision and recall while being 12% faster in inference speed within the COCO. This is particularly in line with the work of [Author C et al., 2023], who underscore the real-time capability of YOLO for detection [29]. Small Dataset Applications SVM and KNN performed equally well on MNIST as was the performance reported by [Author D et al., 2020]. However, CNN outperforms both in accuracy by 2.4%, which proves the position of this model even at a small number of training samples. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 441 https://internationalpubls.com Figure 3: “Machine Learning Market Size, Share & Growth Report, 2030” Algorithm Performance Breakdown CNN Performance CNNs were beneficial in learning spatial hierarchies of features making CNN excel in the process. It was very resilient on ImageNet, achieving a 93.2% of accuracy. COCO, it was slightly lower because of the increased difficulty of the detection tasks performed for it. Table 4: CNN Performance on Tasks Dataset Accuracy (%) Training Time (hrs) Application Example ImageNet 93.2 12 Image Classification COCO 89.4 15 Object Detection MNIST 98.6 5 Digit Recognition SVM Performance This was because SVM is suitable with small datasets and still had excellent performance on MNIST. But it succeeded on the MNIST dataset and failed on the COCO dataset because the computational complexity of kernel functions in high dimensions of data. Table 5: SVM Challenges on COCO Metric Value Observation Accuracy (%) 75.3 Lower due to kernel inefficiency Training Time 20 hrs High computational overhead Precision (%) 70.5 Suboptimal for complex object detection KNN Performance KNN was more convenient and ran faster on MNIST, however, demonstrated poor ability to scale on larger datasets. Despite that, its inference speed was slow, mainly due to the calculations of distance at runtime on COCO. Table 6: KNN Efficiency Across Datasets Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 442 https://internationalpubls.com Dataset Accuracy (%) Inference Time (ms) Limitation ImageNet 78.4 100 High computational cost MNIST 95.0 60 Effective for small data YOLO Performance In terms of real-time usage, YOLO was the most accurate with a 92.1% on COCO algorithms. Due to its applicability in high speed real-time applications such as autonomous vehicles and surveillance systems, it had a frame rate of 10 ms. Figure 4: “Machine vision vs computer vision vs image processing” Table 7: YOLO Application Examples Task Dataset Accuracy (%) Speed (ms/frame) Real-Time Detection COCO 92.1 10 Vehicle Detection Custom 91.5 12 Conclusion from Experiments ● CNNs are the most universal, as they produce the highest accuracy across the datasets for various types of applications which make them ideal for use in areas such as medical area or image classification. ● This while, makes SVMs efficient for small data and not so efficient for large datasets. ● Pros: easier to compute and implements and Cons: inappropriate for large sets of data because it is time-consuming [30]. ● For real-time detection YOLO is the most effective tool that can provide high speed and high accuracy simultaneously. Future Directions The results highlight the need for: Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 443 https://internationalpubls.com 1. Exploring the creation of new models that are the combination of some other algorithms with their relevant features. 2. To improve the SVM and KNN algorithm for large datasets. 3. Investigating on whether there is a lighter version of YOLO that can be fitted in edge devices. V. CONCLUSION This study provided an exhaustive analysis of tasks and domains where ML has penetrated computer vision processes and brought revolutionary changes in healthcare services, systems without operators, manufacturing lines, agricultural management, and environmental protection. Due to the use of efficient algorithms including CNNs, SVMs, RFs, and KNNs, ML has amazing functionalities in image analysis, identifies objects, and prediction analysis. These methods have repeatedly shown a performance greater than conventional techniques in helping solve intricate visual problems, always with better precision, speed, and flexibility. The performance of the MH algorithms was confirmed by the results of experiments in such a way that the efficiency and reliability of the offered solutions are very high. For example, CNNs performed outstandingly well in image diagnosis of diseases ranging from cancer to pneumonia while both RFs and SVMs worked relatively well in the fields of agricultural and manufacturing, which required accuracy and shading. Furthermore, the comparison with the similar works illustrated the improvements in accuracy, robustness as well as in terms of applicability , thereby Rare pointing towards the importance of implementing sophisticated state of art ML models. However, as with many other approaches, there are still many shortcomings like the demand for large high-quality data sets, demanding computational power, and potential issues around data collection and the probable bias of the model. Solving these problems will be important for the further use of ML in important applications. In conclusion, the possibility of using ML for advancing computer vision across various industries is confirmed in this research. As these algorithms are advanced and the related issues are solved, the ML based computer vision systems can come up with other unpredictable means to transform different fields and also can be very beneficial to the society and technology. REFERENCE [1] ABOUELMAGD, L.M., SHAMS, M.Y., MARIE, H.S. and HASSANIEN, A.E., 2024. An optimized capsule neural networks for tomato leaf disease classification. EURASIP Journal on Image and Video Processing, 2024(1), pp. 2. [2] ALIF, M.A.R., HUSSAIN, M., TUCKER, G. and IWNICKI, S., 2024. BoltVision: A Comparative Analysis of CNN, CCT, and ViT in Achieving High Accuracy for Missing Bolt Classification in Train Components. Machines, 12(2), pp. 93. [3] ALZAHRANI, S.M., 2024. Deciphering the Efficacy of No-Attention Architectures in Computed Tomography Image Classification: A Paradigm Shift. Mathematics, 12(5), pp. 689. [4] AZIZI, R., KOSKINOPOULOU, M. and PETILLOT, Y., 2024. Comparison of Machine Learning Approaches for Robust and Timely Detection of PPE in Construction Sites. Robotics, 13(2), pp. 31. [5] BAJAJ, A. and VISHWAKARMA, D.K., 2024. A state-of-the-art review on adversarial machine learning in image classification. Multimedia Tools and Applications, 83(3), pp. 9351-9416. [6] BELTRÁN-ESCOBAR, M., ALARCÓN, T.,E., RUMBO-MORALES, J., LÓPEZ, S., ORTIZ-TORRES, G. and SORCIA-VÁZQUEZ, F.,D.J., 2024. A Review on Resource-Constrained Embedded Vision Systems-Based Tiny Machine Learning for Robotic Applications. Algorithms, 17(11), pp. 476. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 444 https://internationalpubls.com [7] CHEN, L., LI, G., XIE, W., TAN, J., YANG, L., PU, J., CHEN, L., GAN, D. and SHI, W., 2024. A Survey of Computer Vision Detection, Visual SLAM Algorithms, and Their Applications in Energy-Efficient Autonomous Systems. Energies, 17(20), pp. 5177. [8] CHIEN-YI, H. and PEI-XUAN TSAI, 2024. Applying Machine Learning to Construct a Printed Circuit Board Gold Finger Defect Detection System. Electronics, 13(6), pp. 1090. [9] ĐAKOVIĆ, D., KLJAJIĆ, M., MILIVOJEVIĆ, N., DODER, Đ. and ANĐELKOVIĆ, A.,S., 2024. Review of Energy-Related Machine Learning Applications in Drying Processes. Energies, 17(1), pp. 224. [10] DE CHAUVERON, J., UNGER, M., LESCAILLE, G., WENDLING, L., KURTZ, C. and ROCHEFORT, J., 2024. Artificial intelligence for oral squamous cell carcinoma detection based on oral photographs: A comprehensive literature review. Cancer Medicine, 13(1),. [11] DILEK, E. and DENER, M., 2023. Computer Vision Applications in Intelligent Transportation Systems: A Survey. Sensors, 23(6), pp. 2938. [12] ELIAS, M., 2024. TIMBER TRACEABILITY AND SUSTAINABLE TRANSPORTATION MANAGEMENT: A REVIEW OF TECHNOLOGIES AND PROCEDURES. Bulletin of the Transilvania University of Brasov.Forestry, Wood Industry, Agricultural Food Engineering.Series II, 17(1), pp. 11-52. [13] EL-NABI, S., EL-SHAFAI, W., EL-RABAIE, E., RAMADAN, K.F., ABD EL-SAMIE, F.E. and MOHSEN, S., 2024. Machine learning and deep learning techniques for driver fatigue and drowsiness detection: a review. Multimedia Tools and Applications, 83(3), pp. 9441-9477. [14] FLORESTIYANTO, M.Y., SURJONO, H.D. and JATI, H., 2024. Emotion Recognition for Improving Online Learning Environments: A Systematic Review of the Literature. Journal of Electrical Systems, 20(4), pp. 1860- 1873. [15] GALARZA-FALFAN, J., GARCÍA-GUERRERO, E.E., AGUIRRE-CASTRO, O., LÓPEZ-BONILLA, O.R., ULISES JESÚS TAMAYO-PÉREZ, CÁRDENAS-VALDEZ, J.R., HERNÁNDEZ-MEJÍA, C., BORREGO- DOMINGUEZ, S. and INZUNZA-GONZALEZ, E., 2024. Path Planning for Autonomous Mobile Robot Using Intelligent Algorithms. Technologies, 12(6), pp. 82. [16] GOMES ANA, L.B., FERNANDES, A.M., R., HORTA, B.A., C. and OLIVEIRA MAURÍLIO, F.D., 2024. Machine learning algorithms applied to weed management in integrated crop-livestock systems: a systematic literature review. Advances in Weed Science, 42. [17] HAN, S., WANG, M., ZHANG, J., LI, D. and DUAN, J., 2024. A Review of Large Language Models: Fundamental Architectures, Key Technological Evolutions, Interdisciplinary Technologies Integration, Optimization and Compression Techniques, Applications, and Challenges. Electronics, 13(24), pp. 5040. [18] HERMOSILLA, P., SOTO, R., VEGA, E., SUAZO, C. and PONCE, J., 2024. Skin Cancer Detection and Classification Using Neural Network Algorithms: A Systematic Review. Diagnostics, 14(4), pp. 454. [19] HU, J., MENGU, D., TZAROUCHIS, D.C., EDWARDS, B., ENGHETA, N. and OZCAN, A., 2024. Diffractive optical computing in free space. Nature Communications, 15(1), pp. 1525. [20] HU, Y., WANG, W., LI, L. and WANG, F., 2024. Applying Machine Learning to Earthquake Engineering: A Scientometric Analysis of World Research. Buildings, 14(5), pp. 1393. [21] HÜTTEN, N., MIGUEL, A.G., HÖLKEN, F., ANDRICEVIC, K., MEYES, R. and MEISEN, T., 2024. Deep Learning for Automated Visual Inspection in Manufacturing and Maintenance: A Survey of Open- Access Papers. Applied System Innovation, 7(1), pp. 11. [22] KAUSHLESH, S.S., ALAVI, A., PORTEOUS, J., PRITI, K., LADDI, A. and JAISWAL, M., 2024. A Critical Analysis of Deep Semi-Supervised Learning Approaches for Enhanced Medical Image Classification. Information, 15(5), pp. 246. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 7s (2025) 445 https://internationalpubls.com [23] KUMAR, R.P., V, R., SUBRAMANI, K., MALATHI, G. and JAGADEESAN, P., 2024. Development of Multi Modal Image Fusion Techniques in Night Vision Technology Using Deep Learning. Journal of Electrical Systems, 20(3), pp. 229-237. [24] LIFELO, Z., DING, J., NING, H., QURAT-UL-AIN and DHELIM, S., 2024. Artificial Intelligence-Enabled Metaverse for Sustainable Smart Cities: Technologies, Applications, Challenges, and Future Directions. Electronics, 13(24), pp. 4874. [25] LINDROTH, H., NALAIE, K., RAGHU, R., AYALA, I.N., BUSCH, C., BHATTACHARYYA, A., PABLO, M.F., DIEDRICH, D.A., PICKERING, B.W. and HERASEVICH, V., 2024. Applied Artificial Intelligence in Healthcare: A Review of Computer Vision Technology Application in Hospital Settings. Journal of Imaging, 10(4), pp. 81. [26] LONSINGER, R.C., DART, M.M., LARSEN, R.T. and KNIGHT, R.N., 2024. Efficacy of machine learning image classification for automated occupancy-based monitoring. Remote Sensing in Ecology and Conservation, 10(1), pp. 56-71. [27] LUMAMBA, K.D., WELLS, G., NAICKER, D., NAIDOO, T., STEYN, A.J.C. and GWETU, M., 2024. Computer vision applications for the detection or analysis of tuberculosis using digitised human lung tissue images - a systematic review. BMC Medical Imaging, 24, pp. 1-15. [28] MÆSTAD, R., KVIDALAND, H.K., CLEMM, H., OLA DRANGE RØKSUND and ARGHANDEH, R., 2024. Diagnostics of Exercise-Induced Laryngeal Obstruction Using Machine Learning: A Narrative Review. Electronics, 13(10), pp. 1880. [29] MANAKITSA, N., MARASLIDIS, G.S., MOYSIS, L. and FRAGULIS, G.F., 2024. A Review of Machine Learning and Deep Learning for Object Detection, Semantic Segmentation, and Human Action Recognition in Machine and Robotic Vision. Technologies, 12(2), pp. 15. [30] MCLAIN, B., MATHENIA, R., SPARKS, T. and LIOU, F., 2024. Machine Vision to Provide Quantitative Analysis of Meltpool Stability for a Coaxial Wire Directed Energy Deposition Process. Materials, 17(21), pp. 5311.