typo fixes

paper/paper.md

@@ -38,9 +38,9 @@ Integration of machine learning models with mobile devices presents a promising
 # Workflow
 The app uses the garbage classification model obtained by applying transfer learning approach to the  garbage dataset [@suman_kunwar_2023]. This dataset consists of 10 categories, including plastic, metal, glass, biological, paper, battery, trash, cardboard, shoes and clothes. Few sample images from the dataset are shown in \autoref{fig:sample_images} and  the count of each classes are shown in \autoref{fig:garbage_dataset}.
 
-![Sampel images from dataset \label{fig:sample_images}](sample_images.png)
+![Sample images from dataset \label{fig:sample_images}](sample_images.png)
 
-![Garbage Dataset\label{fig:garbage_dataset}](garbage_dataset.png)
+![Count of images in each class of Garbage Dataset\label{fig:garbage_dataset}](garbage_dataset.png)
 
 With uneven number of images in each classes, random undersampling [@LIU2020105292] is employed to balance uneven class sizes by excluding some data from the larger dataset along with image augmentation methods to address class imbalance. The dataset is divided into 3 sets: train (80%), validation (10%), and test. The train set trains the model, the validation set tunes parameters, and the test set evaluates accuracy on new data.
 
@@ -52,15 +52,15 @@ The model was trained with Tesla T4 GPU and uses EfficientNetV2 [@tan2021efficie
 
 The accuracy of the trained model at 20 epoch was found to be 96%. \autoref{fig:confusion_matrix} shows the confusion matix of the model.
 
-![Confusion Matrix\label{fig:confusion_matrix}](confusion_matrix.png)
+![Confusion Matrix of Model\label{fig:confusion_matrix}](confusion_matrix.png)
 
 The hyperparameters were optmised using [optuna](https://optuna.org/) to create more accurate results. \autoref{fig:hyperparameters_optimization} shows the hyperparameter optmization over each trial. This optmized parameters were then fed into the model for training purpose. We found that the model performance was increased slighlty and the new accuracy is 96.41%. \autoref{fig:test_results} shows the test results with various sample test images.
 
 ![Hyperparameters Optimization\label{fig:hyperparameters_optimization}](optimization.png)
 
 
 
-![Test Results\label{fig:test_results}](test_results.png)
+![Test results of sample images\label{fig:test_results}](test_results.png)
 
 The classification model is then converted into a lite format, such as [TFLite](https://www.tensorflow.org/lite/guide), which enables it to be used on mobile devices with limited resources. This format allows for fast loading times, smaller size, and compatibility with various programming languages and platforms. \autoref{fig:deep_waste_app_workflow} describes the overall workflow of the app.