Intelligent ripening classification of nendran banana(Musa spp.)using convolutional neural network(CNN) (Record no. 291386)
[ view plain ]
| 000 -LEADER | |
|---|---|
| fixed length control field | 04869nam a22001937a 4500 |
| 082 ## - DEWEY DECIMAL CLASSIFICATION NUMBER | |
| Classification number | 630.31 |
| Item number | ARU/IN PG |
| 100 ## - MAIN ENTRY--PERSONAL NAME | |
| Personal name | Arunima, P L |
| 245 ## - TITLE STATEMENT | |
| Title | Intelligent ripening classification of nendran banana(Musa spp.)using convolutional neural network(CNN) |
| 260 ## - PUBLICATION, DISTRIBUTION, ETC. (IMPRINT) | |
| Place of publication, distribution, etc | Vellayani |
| Name of publisher, distributor, etc | Department of Agricultural Statistics, College of Agriculture |
| Date of publication, distribution, etc | 2024 |
| 300 ## - PHYSICAL DESCRIPTION | |
| Extent | 104p. |
| 502 ## - DISSERTATION NOTE | |
| Dissertation note | MSc |
| 520 3# - SUMMARY, ETC. | |
| Abstract | The research work entitled “An intelligent ripening classification of Nendran banana (Musa spp.) using Convolutional Neural Network (CNN)” was carried out at the College of Agriculture, Vellayani during the period 2021-2023. The objectives were the development of ripe banana identification algorithm using Convolutional Neural Network (CNN). Evaluation of the performance of classification algorithms, Random Forest and K-Nearest Neighborhood (KNN). Unripe bananas were collected from the selected fields randomly and these bananas were subjected to a typical day/night storage cycle. An android device with triple camera of 50 MP, 8 MP and 2 MP is used for the image acquisition under white light illumination. Together a dataset of 4320 images belonging to 4 classes: stage 1- unripe, stage 2- medium ripe, stage 3- ripe and stage 4- overripe were gathered. Image cropping, image resizing and data augmentation were employed to pre-process the image dataset. The dataset was split into train-test split in the ratio 80:20 and attempted to utilize the pre-existing deep learning models VGG16 (Visual Geometry Group 16), VGG19, InceptionV3, ResNet50 (Residual Neural Network 50) and EfficientNet B0 to carry out image classification. VGG16 demonstrated a testing accuracy of 86 percent, whereas VGG19 and InceptionV3 achieved accuracies of 83 percent and 84 percent, respectively, on the test set. While ResNet50 and EfficientNetB0 achieved relatively modest levels of accuracy on the test data, specifically 24 percent and 23 percent, respectively. Hence, an endeavor was made to create a novel CNN model through the optimization of hyperparameters. Tuned the hyperparameters of the model before training and testing of data. The hyperparameters, including batch size, number of epochs, dropout rate, optimizer and learning rate, were selected based on the evaluation of the Area Under Curve (AUC) of the Receiver Operating Characteristic (ROC) curve. Proposed a 9- layer CNN architecture featuring hyperparameters such as a batch size of 32, number of epochs set to 100, a dropout rate of 0.2 and the Adaptive Moment Estimation (Adam) as the optimizer. The learning rate was set at 0.002. Additionally, the loss function employed was the categorical cross-entropy, while the activation functions used were Softmax and ReLU (Rectified Linear Unit). Model evaluation using the train set acquired an accuracy of 95 percent and 93 percent accuracy on the validation set. The model performed well with the unseen test data and achieved an accuracy of 95 percent. Multivariate analysis of variance (MANOVA) was performed over the four ripeness classes for the variables Red (R), Green (G) and Blue (B) and it is observed that there is a significant difference (p<0.01) between the ripeness classes based on the average RGB values. So, we assumed that average RGB values can also be used as the factors for classifying the banana based on the ripeness. In order to implement this, two ML algorithms K-Nearest Neighbor (KNN) and Random Forest (RF) were used. KNN and RF attained an accuracy of 65 percent and 64 percent respectively. It can be further improved with a large number of images as it is highly beneficial with very low execution time. It is also clear that average RGB values can be used to find the significant difference between the ripening stages of bananas. A web application was developed to demonstrate Nendran banana ripeness classification using images entitled ‘Banana Ripeness Identification App’ with the proposed CNN model where the users can test the accuracy of the model by uploading their images. The application was developed using the web development platform RShiny in conjunction with the open-source language R and its integrated development environment RStudio. Example datasets were also provided to test the model. A ripening classification model was developed for Nendran banana using CNN. On comparing the three models CNN, KNN and Random Forest, it is concluded that the proposed CNN model classifies the Nendran bananas more accurately with an accuracy of 95 percent. Moreover, the proposed 9-layer CNN model is more accurate than the existing deep learning models VGG16, VGG19, Inception V3, ResNet50 and EfficientNet B0. |
| 650 ## - SUBJECT ADDED ENTRY--TOPICAL TERM | |
| Topical term or geographic name as entry element | Agricultural Statistics |
| 650 ## - SUBJECT ADDED ENTRY--TOPICAL TERM | |
| Topical term or geographic name as entry element | Nendran Banana |
| 650 ## - SUBJECT ADDED ENTRY--TOPICAL TERM | |
| Topical term or geographic name as entry element | Musa spp. |
| 650 ## - SUBJECT ADDED ENTRY--TOPICAL TERM | |
| Topical term or geographic name as entry element | Neural network |
| 700 ## - ADDED ENTRY--PERSONAL NAME | |
| Personal name | Pratheesh P Gopinath(Guide) |
| 942 ## - ADDED ENTRY ELEMENTS (KOHA) | |
| Source of classification or shelving scheme | |
| Item type | Theses |
| Not for loan | Collection code | Permanent location | Current location | Shelving location | Date acquired | Full call number | Barcode | Date last seen | Koha item type |
|---|---|---|---|---|---|---|---|---|---|
| Not For Loan | Thesis | KAU Central Library, Thrissur | KAU Central Library, Thrissur | Theses | 2024-07-03 | 630.31 ARU/IN PG | 176030 | 2024-07-03 | Theses |