<p>The increased usage of PET bottles has resulted in a notable surge in plastic waste, prompting the implementation of automated reverse vending machines (RVMs) as a potential solution. However, the presence of contaminants in the bottles impedes the recycling process. Once RVMs gain the capability to automatically inspect contamination, the rate of PET material recycling will be enhanced. In this paper, four inspection algorithms were considered: logistic regression (LR), fully connected neural networks (FCNN), convolutional neural networks (CNNs), and CNN via transfer learning (CNN-TL). To evaluate the efficacy of these technologies, clean and contaminated PET bottles were subjected to an experimental analysis. In order to prepare the data for training the AI agent, the RGB colors at the bottom of 253 bottles were measured, and more than 120 images for each class were also captured. The LR and FCNN algorithms exhibited lower performance, with accuracy rates below 80 % when utilizing the RGB numerical data. In contrast, the image-based CNN and CNN-TL algorithms demonstrated almost 100 % and 98 % accuracy, respectively. As the 100 % accuracy can be obtained by overfitting, in this study, the CNN-TL via the DenseNet121 model was selected. Future research endeavors will explore the creation of the CNN-TL in conjunction with physical sensor data to prevent a missing detection of the contaminated bottle. This will further enhance the accuracy of the inspection process through the use of fusion algorithms.</p>