ICAICTA 2024 6 hal.docx

Comparison of ResNet50 and MobileNetv2 Methods for Agricultural Crop Classification *Abstract*---Agriculture is an important sector in economic development. The application of technology, especially machine learning, is needed to manage agricultural data. This research aims to develop a classificat...

Comparison of ResNet50 and MobileNetv2 Methods for Agricultural Crop Classification *Abstract*---Agriculture is an important sector in economic development. The application of technology, especially machine learning, is needed to manage agricultural data. This research aims to develop a classification model by utilizing deep learning models, namely ResNet50 and MobileNetV2, for Four agricultural crops: rice, corn, cassava, and sugarcane. Both models were trained on the same plant dataset, and hyperparameters such as image size and number of epochs were tuned. The research results show that proper hyperparameters can significantly improve accuracy. The ResNet50 and MobileNetV2 models achieve accuracies of 63.83% and 87.23%, respectively. This research proves that the use of deep learning models and hyperparameter tuning can effectively improve plant type classification performance. Keywords--- classification, hyperparameters, MobileNetv2, ResNet50 Introduction ============ Convolutional Neural Network (CNN) is a deep learning algorithm used to recognize digital images. As a digital image recognition algorithm, this algorithm is very effective for classification and has many models \[1\]. Digital image recognition has great potential for facial recognition \[2\], pattern identification, and classification. The CNN structure consists of many layers, including convolution, pooling, and fully connected layers. CNN works by extracting important features in input data, reducing dimensions, and classifying data. CNN models that can accommodate integration needs into mobile applications include MobileNetV2 and ResNet50 \[3\]. In CNN, ResNet50 is a model that has higher accuracy in image recognition than other models, such as VGG16 and VGG19 \[1\]\[4\]. The Renet50 model can effectively solve the problem of gradient disappearance and network degradation by using connections in a deep CNN model using a residual structure \[5\] \[6\]. MobileNet is a CNN model that is effectively used as a classification model whose use is aimed at mobile devices, especially the MobileNetV2 model. Image testing results using MobileNetV2 have a higher level of accuracy and shorter training time than the MobileNetV1 model \[7\]. The MobileNetV2 model is ideal for implementation on mobile devices with limited resources. In the context of plant type classification, selecting the right CNN model and optimizing hyperparameters is very important to improve accuracy and efficiency \[8\]. Hyperparameters such as learning rate, image size, and number of epochs significantly influence model performance. Setting the right hyperparameters can increase model accuracy \[9\]. This research aims to develop a model and analyze hyperparameters for plant type classification using two well-known CNN models, namely ResNet50 and MobileNetV2. The main objective of this research is to answer the question of how optimal hyperparameter settings in these two models can improve the accuracy of plant type classification. This question is important because accurate classification has broad applications in precision agriculture, crop monitoring, and technology-based agricultural management systems. This research is expected to make a significant contribution to the field of digital image recognition and plant type classification by providing an in-depth analysis of the influence of hyperparameters on the performance of ResNet50 models and MobileNetV2. The objective of this research is to deliver a more effective and efficient classification system, which will ultimately support innovation in the field of agricultural technology. Related Works ============= Classification using two models derived from CNN, such as ResNet50 and MobileNetV2, is often used and compared regarding accuracy and processing speed. However, no one has yet classified several types of agricultural crops using this model. Classification using the ResNet50 and MobileNetv2 algorithm models has been used, but the research objects or data used in this research are four breeds of cattle endemic to Indonesia \[3\]. Previous study examined medicinal plant classification using the ResNet50 model, which utilizes the depth of the ResNet50 layers to be trained on a comprehensive dataset of medicinal leaf images and uses transfer learning technology \[10\]. This study used 80 types of medicinal plants and the ResNet50 classification model was able to achieve accuracy of up to 99.84%. However, the data used to train the model was only an image of a single leaf, so the model may not be able to recognize medicinal plants that are still alive or intact. ResNet50 and GoogleNet were modified as Classification Similarity Network (CSN) for image fusion with medical images \[11\]. In a study on developing the ResNet50 model with the transfer learning method used to speed up the development process, the results reached 73.33% for accuracy where the model implemented Python and TensorFlow \[12\]. ResNet50 with SVM achieves an accuracy of up to 99% and ResNet50 using the Softmax model achieves an accuracy of 98.8% \[13\]. Another study shows that adding layers to the ResNet50 architecture can increase the accuracy of wheat leaf disease classification, reaching up to 98.44% \[14\]. ResNet50 was used to detect and classify rotten fruit, resulting in a validation accuracy of 98.89% with a classification time of around 0.2 seconds per image. Although the results are good on monotonous backgrounds, accuracy decreases to 85% on images with complex backgrounds \[15\]. Although efficient and fast ResNet50, thanks to its pre-trained architecture, is not always suitable for all datasets, utilizing ResNet50 with transfer learning can provide an accuracy of 93.5% \[15\]. The ResNet50 algorithm using the Cyclical Learning Rate produces a model with an accuracy of up to 95%. In this research, the setup with a small CLR and no dropout has produced good accuracy and loss results \[16\]. Classification using MobileNetV2 is very effective, in a lotus plant classification study the pre-trained MobileNetV2 model was able to classify names in the lotus class with the highest accuracy of 99.5% \[17\]. In another study, the MobileNetV2 model achieved stable accuracy of 95.75%, 96.74%, and 96.23% on datasets 1, 2, and 3 respectively in fruit image classification \[18\]. In addition, MobileNetV2, with its lightweight network architecture and inverted residual module with linear bottleneck, is increasingly popular in various deep learning applications due to its memory efficiency and high accuracy \[19\]. In the melanoma classification study, MobileNetV2 was used as a base model by adding a global pooling layer and two fully-connected layers, achieving an accuracy of up to 85% on the ISIC-Archive dataset, showing its superiority over other models such as ResNet50V2 and InceptionV3 \[19\]. In the fruit image classification study, MobileNetV2 was modified to TL-MobileNetV2 by adding five special layers at the head of the model. Using transfer learning, TL-MobileNetV2 achieves 99% accuracy, 3% higher than the original MobileNetV2, and shows improved performance compared to AlexNet, VGG16, InceptionV3, and ResNet \[20\]. MobileNetV2, designed for mobile applications, excels in accuracy on fruit datasets without pre-processing techniques, demonstrating the superiority of its architecture in computer vision tasks \[20\]. In other research, MobileNetV2 was combined with the Convolutional Block Attention Module (CBAM) to form the CBAM-MobileNetV2 model, which was used to classify citrus huanglongbing disease. Using transfer learning and parameter fine-tuning, CBAM-MobileNetV2 achieves 98.75% accuracy, higher than the MobileNetV2, Xception, and InceptionV3 models. This study shows that combining CBAM with MobileNetV2 improves the model's ability to capture semantic information and improves classification accuracy \[21\]. CBAM-MobileNetV2 has the disadvantages of higher model complexity and longer training time compared to the original MobileNetV2, which also leads to greater memory usage. In addition, the risk of overfitting increases and implementation becomes more difficult due to the addition of attention modules \[21\]. In another study, the MobileNetV2-UNET model was used to segment leaves of guava, potato and jamun plants with an accuracy of 96.38%. However, its accuracy is still lower than that of the hybrid Plant Species Detection Stacking Ensemble Deep Learning Model (PSD-SE-DLM), which achieved 99.84% accuracy for plant classification \[22\]. The MobileNetV2 model is used for the classification of traditional Indian medicinal flowers with the highest accuracy reaching 98.23%, showing its superiority in flower images that have different environmental backgrounds \[23\]. In another study, the MobileNetV2 model was tested with several different parameters to achieve the best accuracy, producing an accuracy of 75% on training data and 67% on validation data in ornamental plant classification. This result is better than the built CNN model, even though the accuracy level is still relatively lower than expected \[24\]. MobileNetV2 with hyperparameters was used for grape leaf disease classification and achieved 99.94% accuracy after hyperparameter optimization using a hyperband strategy, showing superiority over traditional CNN models in computational efficiency and memory usage \[25\]. In another study, MobileNetV2 was used for seed classification and achieved 98% accuracy on training data and 95% on testing data, demonstrating the superiority of a simple and efficient architecture in memory usage compared to traditional CNN models \[26\]. Proposed Model ============== Datasets -------- The dataset was obtained from Kaggle with search categories in the form of images of four types of plants that are the subject of classification. The images used are in jpg format and there are a total of 245 images, 80% of which are used for training and 20% for testing. Fig 1. Images of Rice, Corn, Cassava, and Sugarcane ResNet50 Model Architecture --------------------------- ResNet50 was introduced by Microsoft Research in 2015 and won the ImageNet image recognition competition which demonstrated the superiority of ResNet in image recognition \[7\] \[8\]. This ResNet model consists of several Residual blocks that overcome the problem of degradation in deep networks by using shortcut connections. The model receives an input image of some size in pixels, which goes through several stages of convolution, batch normalization, and ReLU activation. Residual blocks include identity blocks and convolutional blocks with multilevel filters (64, 128, 256). Shortcut connections add together the original input with the convolution output, helping to retain information. After going through pooling and flattening layers, the model uses Dense and Dropout layers before producing output with a softmax activation function for multi-class classification. The model was compiled with the Adam optimizer and categorical crossentropy loss, then trained for 150 epochs. Testing data is used for evaluation, measuring accuracy, precision, recall, f1 score, and creating a confusion matrix. Training and evaluation results are saved in text files, while accuracy and loss graphs are visualized for monitoring model performance. Fig 2. ResNet50 Architecture MobileNetV2 Model Architecture ------------------------------ MobileNetV2 was proposed as a model for the first time by the Google Research team in 2019 as an improvisation of the MobileNetV2 model to effectively maximize accuracy by considering resource limitations on mobile devices \[7\] \[14\]. This MobileNetV2 model uses additional layers for classification, such as GlobalAveragePooling2D, Dropout, and Dense, which ends with a softmax layer for class prediction. The base layers of MobileNetV2 are frozen to prevent updates during training, while additional layers are trained using the training data. The model is compiled with the Adam optimizer and sparse categorical crossentropy loss. The training process was carried out for 150 epochs with a batch size of 32. Training and testing data were normalized to the range \[0, 1\] before use. After training, the model is evaluated using test data to calculate metrics such as accuracy, precision, recall, and f1 score. Training and evaluation results are saved in text files, while accuracy and loss graphs are visualized to monitor model performance. The trained model is also saved in 'keras' format. Fig 3. MobileNetV2 Architecture Hyperparameters --------------- Hyperparameters are used for the learning or testing process with two types of parameters including image size and epoch. There are several image sizes tested including 32x32, 50x50, 75x75, 100x100, 224x224, and 448x448 pixels. The training process is carried out during different epochs, including 50 epochs, 100 epochs, and 150 epochs. Each combination of image size and number of epochs is tested to see its effect on model performance. The evaluation results of each hyperparameter combination are compared using metrics such as accuracy, precision, recall, and f1 score. The best performance is recorded and used to determine the optimal hyperparameter settings that provide the most accurate and efficient results. Results and Discussion ====================== Based on the test results, it can be seen that image size significantly impacts the performance of both models. Larger image sizes tend to provide higher accuracy. It can be seen on MobileNetV2 with an image size of 448x448, achieving the highest accuracy of 87.23% with 50 and 100 epochs. Fig 4 shows the testing result of MobileNetV2. Fig 4. Testing result of MobileNetV2 On ResNet50, an image size of 75x75 provides better results than other image sizes, with accuracy reaching 63.83% at 150 epochs. However, the 100x100 image size shows a decrease in performance at higher epochs, indicating that ResNet50 may not manage higher image resolutions effectively in this setting. Fig 5 shows testing result of ResNet50. Fig 5. Testing result of ResNet50 Effect of the Number of Epochs on Model Performance --------------------------------------------------- The number of epochs also has a significant influence on model performance. In general, increasing the number of epochs tends to increase model accuracy up to a certain point, after which accuracy can decrease or stabilize. For example, on MobileNetV2 with an image size of 224x224, the highest accuracy was achieved at 50 epochs (78.72%) and tended to be stable at 100 and 150 epochs. On ResNet50, increasing the number of epochs from 50 to 100 and 150 shows a general increase in accuracy, but the best results remain at an image size of 75x75 with 150 epochs. Effect of Image Size and Number of Epochs on Precision, Recall, and F1 Score ---------------------------------------------------------------------------- Apart from accuracy, precision, recall, and F1 score also provide important insights into model performance. Precision indicates the proportion of positive predictions that are truly positive. MobileNetV2 achieved the highest precision of 89.29% on an image size of 448x448 with 50 epochs, while ResNet50 achieved the highest precision of 71.55% on an image size of 75x75 with 50 epochs. The higher precision in MobileNetV2 indicates that this model is more effective in reducing false positive predictions. Recall measures the proportion of total positive data that is actually detected by the model. MobileNetV2 achieved the highest recall of 87.09% on an image size of 448x448 with 50 epochs, while ResNet50 achieved the highest recall of 64.29% on an image size of 75x75 with 150 epochs. The higher recall on MobileNetV2 indicates that this model is better at detecting all positive data in the dataset. F1 Score is the harmonic mean of precision and recall, which provides a comprehensive picture of the balance between precision and recall. MobileNetV2 achieved the highest F1 score of 87.61% on an image size of 448x448 with 50 epochs, while ResNet50 achieved the highest F1 score of 59.44% on an image size of 75x75 with 150 epochs. The higher F1 score on MobileNetV2 shows that this model has a better balance between precision and recall. Comparison between MobileNetV2 and ResNet50 ------------------------------------------- Overall, MobileNetV2 shows better performance than ResNet50 in image classification tasks on this dataset. Apart from accuracy, MobileNetV2 also excels in terms of precision, recall and F1 score. MobileNetV2 achieved the highest accuracy of 87.23%, the highest precision of 89.29%, the highest recall of 87.09%, and the highest F1 score of 87.61% on images measuring 448x448 with 50 epochs. Meanwhile, ResNet50 achieved the highest accuracy of 63.83%, the highest precision of 71.55%, the highest recall of 64.29%, and the highest F1 score of 59.44% on images measuring 75x75 with 150 epochs. MobileNetV2 also shows better performance stability at larger image sizes and higher number of epochs, compared to ResNet50 which tends to show performance fluctuations. Conclusion ---------- Based on the test and analysis results, it can be concluded that MobileNetV2 is superior to ResNet50 in image classification tasks on this dataset. Apart from accuracy, MobileNetV2 also shows better performance in terms of precision, recall and F1 score. Larger image sizes and optimal number of epochs make a positive contribution to improving model performance. This research provides valuable insights into the selection of model architecture and hyperparameter settings for image classification tasks. evaluation ========== Evaluation using the Confusion matrix provides a detailed picture of the model's classification performance by showing the number of correct and incorrect predictions made for each class. The following is a confusion matrix analysis for MobileNetV2 and ResNet50 based on their respective best configurations. --- --- a b --- --- Fig 6. Confusion matrix a. 448x448 image size and 50 epoch b. 75x75 image size and 150 epoch This confusion matrix analysis was carried out for MobileNetV2 and ResNet50 based on their respective best configurations. In the confusion matrix there are four key values: True Positive (TP), False Positive (FP), False Negative (FN), and True Negative (TN). The purpose of the evaluation in this study was to measure the performance of two different architectural models, ResNet50 and MobileNetV2, in plant image classification by varying the image size parameters (32x32, 50x50, 75x75, 100x100, 224x224, and 448x448 pixels) and the number of epochs (50, 100, and 150). This test monitors accuracy, precision, recall and F1 score using the formula: \ [\$\$Accuracy\\ = \\ \\frac{TP + TN}{TP + TN + FP + FN}\$\$]{.math.display}\ \ [\$\$Precision = \\frac{\\text{TP}}{TP + FP}\$\$]{.math.display}\ \ [\$\$Recall = \\ \\frac{\\text{TP}}{TP + FN}\$\$]{.math.display}\ \ [\$\$F1\\ score = \\frac{2\\ \\times Recall\\ \\times Precision}{Recall + Precision}\$\$]{.math.display}\ Performance comparison between MobileNetV2 and ResNet50 shows that MobileNetV2 has better accuracy and balance between precision and recall. The larger image size and optimal number of epochs in MobileNetV2 provide better results compared to ResNet50. The results of this study confirm that MobileNetV2 is a better choice for image classification tasks on this dataset. For practical implementation, it is recommended to use MobileNetV2 with a larger image size and a sufficient number of epochs to achieve optimal accuracy and balance. The conclusion of the evaluation using the confusion matrix shows that MobileNetV2 not only achieves higher accuracy but also performs better precision, recall, and F1 score, providing important guidance in model selection and hyperparameter settings for similar image classification tasks in the future. Conclusion and Recommendation ============================= Based on the experiments and analysis carried out in this research, it can be concluded that MobileNetV2 consistently outperforms ResNet50 in performance on rice, corn, cassava and sugarcane image datasets. MobileNetV2 shows higher accuracy, precision, recall, and F1 score than ResNet50, especially at larger image sizes. The size of 448x448 with 50 epochs produces the highest accuracy of 87.23%, while ResNet50 achieves the best results at 150 epochs with an image size of 75x75. For recommendations, it is recommended that MobileNetV2 be used for image classification with a larger image size and an optimal number of epochs around 50 to achieve the best performance. Further research needs to be done to explore combinations of hyperparameters such as learning rate and batch size. Additionally, data augmentation can increase the amount of training data and help the model generalize better. For further validation, testing on larger and more diverse datasets is also recommended, as well as investigating other models such as EfficientNet and DenseNet to evaluate potential performance improvements in this classification task. By following these recommendations, it is hoped that improvements in model performance in image classification can be achieved, as well as better contributions to research and practical applications in the fields of image recognition and computer vision. ##### Acknowledgment Thanks to research and community service institution, Universitas Pendidikan Ganesha for the fund given. And also, KGEO Thailand for the research collaboration. ##### References \[1\] Sushma, L. and Laksmi, K. P. "An Analysis of Convolution Neural Network for Image Classification using Different Models", International Journal of Engineering Research & Technology, vol. 9, no. 10, pp: 629-637, 2020. doi: \[2\] Budiman, A., Fabian, Yaputera, R. A., Achmad, S., and Kurniawan, A., "Student attendance with face recognition (LBPH or CNN): Systematic Literatur Rewiew", *Elsevier,* vol. 216, pp. 31-38, 2023, doi: 10.1016/j.procs.2022.12.108. \[3\] Fikri, A., and Murni, A., "New Generation Indonesian Endemic Cattle Classification MobileNetV2 and ResNet50", *Jurnal Ilmiah Teknik Elektro Komputer dan Informatika*, vol. 9, no. 4, pp. 941-950, 2023. doi:10.26555/jiteki.v9i4.26659 \[4\] Astuti, R. P., Rachmawati, E., Siregar, S., Sardi, I. L., Fahmi, A., Agustian, Y., Putra, A. C. A. Y., and Daffa, F., "Vegetation classification algorithm using convolutional neural network ResNet50 for vegetation mapping in Bandung district area", *Jurnal Infotel*, 2022, Vol.14, No.2, doi: https://doi.org/10.20895/infotel.v14i2.756 \[5\] Bai, Z.; Zhu, R.; He, D.; Wang, S.; Huang, Z., "Adulteration Detection of Pork in Mutton Using Smart Phone with the CBAM-Invert-ResNet and Multiple Parts Feature Fusion". *Foods* 2023, 12, 3594. https://doi.org/10.3390/ foods12193594 \[6\] Simbolon, H. F. S., "Classification bootstrap ResNet50 for potato diseases datasets", *Jurnal Mantik*, vol. 7, no. 4, pp. 3334-3341, 2024. doi: \[7\] Dong, K., Zhou, C., Ruan, Y., and Li, Y., "MobileNetV2 model for image classification", *Information Technology and Computer Application,* vol. 2, pp. 476-480, 2020. doi: https://doi.org/10.1109/ITCA52113.2020.00106 \[8\] Prottasha, S. I., Hossain, A. B. M. K., Rahman, Z., Reza S. M. S., and Hossain, D. A., "Identification of Various Rice Plant Diseases Using Optimized Convolutional Neural Network", *International Journal of Computing and Digital Systems,* vol. 12, no. 1, pp. 1539-1551, 2022. doi:https://dx.doi.org/10.12785/ijcds/1201124 \[9\] Raiaan, M. A. K., Sakib, S., Fahad, N. M., Mamun, A. A., Rahman, A., Shatabda, S., and Mukta, S. H., "A systematic review of hyperparameter optimization techniques in Convolutional Neural Networks", *Elsevier,* vol. 11, pp. 1-32, 2024. doi: https://doi.org/10.1016/j.dajour.2024.100470 \[10\] Mantravadi, B. S., Vintha, V. V. S. A., Logisa, S., Doki, R., and Mummana, P., "A Robust Approach to Medicinal Plant Classification through the ResNet50 Model", *Journal Of Technology*, vol. 14, no. 4, pp. 27-31, doi: \[11\] Sreedhar, P. S. S., and Nandhagopal, N., "Classification Similiarity Network Model for Image Fusion Using ResNet50 and GoogLeNet", *Tech Science Press,* 2022. doi: 10.32604/iasc.2022.020918. \[12\] Faisal, K. M., Hossien, E., Islam, M. T., "Identification and Classification of Rice Disease Using ResNet 50", *Internasional Journal of Scientific & Engineering Research,* vol. 13, 2022. \[13\] Alwan D. S., and Naji, M. H., "Rice Disease Classification by Residual Netwrok 50 (RESNET50) and Support Vector Machine (SVM) Modeling", *Journal of Kufa for Mathematics and Computer,* vol. 10, no. 1, pp. 96-101, 2023. doi: http://dx.doi.org/10.31642/JoKMC/2018/100114 \[14\] Usha Ruby A., George Chellin Chandran, J., Chaithanya, B. N., Swasthika Jain, T. J., and Patil, R., "Wheat Leaf Disease Classification Using Modified ResNet50 Convolutional Neural Network Model", *Research Square,* 2022. doi: \[15\] Chai C. Foong, Ghong K. Meng, and Lim L. Tze, "Convolutional Neural Network based Rotten Fruit Detection using ResNet50", *Control and System Graduate Research Colloquium,* 2021, doi: 10.1109/ICSGRC53186.2021.9515280 \[16\] Kumar, R., Chug, A., Singh, A. P., and Singh, D., "Evaluation of Deep learning based ResNet-50 for Plant Disease Classification with Stability Analysis", *International Conference on Intelligent Computing and Control Systems,* 2022, doi: 10.1109/ICICCS53718.2022.9788207 \[17\] Rachburee, N., and Punlumjeak, W., "Lotus species classification using transfer learning based on VGG16, ResNet152V2, and MobileNetV2", *IAES Internasional Journal of Artificial Intelligence,* vol. 11, no. 4, pp. 1344-1352, doi: 2022. 10.11591/ijai.v11.i4.pp1344-1352. \[18\] Shahi, T. B., Sitaula, C., Neupane, A., and Guo, W., "Fruit classification using attention-based MobileNetV2 for industrial applications", *PloS ONE,* vol. 12, no. 2, 2022. Doi: https://doi.org/10.1371/journal.pone.0264586 \[19\] Indraswari, R., Rokhana, R., and Herulambang, W.,, "Melanoma image classification based on MobileNetV2 network", *Elsevier,* pp. 198-207, 2022. doi: 10.1016/j.procs.2021.12.132 \[20\] Gulzar, Y. Fruit Image Classification Model Based on MobileNetV2 with Deep Transfer Learning Technique. *Sustainability* 2023, 15, 1906. https://doi.org/ 10.3390/su15031906 \[21\] Dou, S.; Wang, L.; Fan, D.; Miao, L.; Yan, J.; He, H. "Classification of Citrus Huanglongbing Degree Based on CBAM-MobileNetV2 and Transfer Learning". *Sensors* 2023, 23, 5587. https://doi.org/10.3390/ s23125587 \[22\] Rashid, J., Khan, I., Abbasi, I. A., Saeed, M. R., Saddique, M., and Abbas, M., "A Hybrid Deep Learning Approach to Classify the Plant Leaf Species", *Computers, Materials & Continua,* vol. 76, no. 3, pp. 3897-3920, 2023, doi: 10.32604/cmc.2023.040356 \[23\] Bipin Nair, B. J., Arjun, B., Abhishek, S., Abhinav N. M., and Madhavan V., "Classification of Indian Medicinal Flowers using MobileNetV2", *IEEE,* pp. 1512-1518, 2024. \[24\] Parjito, Ulum, F., Muludi, K., Abidin, Z., Alma, R. M., and Permata, "Classification of Ornamental Plants with Convolutional Neural Networks and MobileNetV2 Approach", *International Seminar on Machine Learning, Optimization, and Data Science,* 2022, doi: : 10.1109/ISMODE56940.2022.10180988 \[25\] Hoang-Tu Vo, Kheo Chau Mui, Nhon Nguyen Thien, Phuc Pham Tien, and Huan Lam Le, "Optimizing Grape Leaf Disease Identification Through Transfer Learning and Hyperparameter Tuning", *Internasional Journal of Advanced Computer Science and Applications,* 2024, vol. 15, no. 2, pp. 922-931 \[26\] Hamid, Y., Wani, S., Sommro, A. B., Alwan, A. A., Gulzar, Y., "Smart Seed Classification System based on MobileNetV2 Architecture", *International Conference on Computing and Information Technology,* 2022, doi: 10.1109/ICCIT52419.2022.9711662