3.1 Preamble
This chapter presents the results of the Machine Learning model developed to detect plastic bottle wastes and the subsequent differentiation of plastic bottles from glass bottles based on weight. The implementation of the model using an embedded system for a prototype will also be discussed, along with the design outcome for the inscriptions on biodegradable and non-biodegradable waste bins. The chapter offers insights into the model's effectiveness, practical deployment, and the potential aftermath for improving waste management practices through clear identification and segregation of different types of waste via the execution of the project.
3.2 Accuracy of the Model and Confusing Matrix Evaluation
With an accuracy of 99.5%, the Machine Learning model presented exceptional performance in detecting plastic bottle wastes. The confusion matrix (Figure 3.1) revealed that, when rounded to one decimal place, 1.0 of the test data fell within the true positive region, indicating the model's precision in accurately differentiating plastic bottles from other materials.
3.3 Precision – Recall Curve
The Precision-Recall Curve (Figure 3.2) shows how well a classification model performs at different thresholds. In this case, the result "0.995 [email protected]" indicates that the model achieved a very high level of accuracy (0.995) and average precision (mAP) when using a threshold of 0.5 for all categories. This suggests that the model was able to classify the data points with a high proportion of correct positive predictions and very few incorrect ones.
3.7 Results of the Trained Model on Unseen Dataset
The results of the trained model on an unseen dataset, as shown in Figure 3.3, are crucial as they demonstrate its performance on new or unseen data. By using evaluation metrics like accuracy, precision, and recall on this dataset, the effectiveness of the model in real-life scenarios can be assessed.
3.8 Results of the Weight Differentiation of Plastic Bottles from Glass Bottles
Using Table from Appendix D for 100 Samples,
3.9 Inscription Design Outcome
Having unique inscription designs for biodegradable (Figure 3.4) and non-biodegradable wastes (Figure 3.5) is crucial because it makes it clear and easy to tell them apart and sort them correctly. When there are clear symbols or labels that everyone can recognize for each type of waste, people can quickly identify the right bins or containers to use for disposal. This helps everyone to manage waste effectively and prevents mixing biodegradable and non-biodegradable materials together, which could harm the environment.
3.10 Discussion
3.10.1 The Machine Learning Model
The improvement in the model's accuracy for this study compared to a previous article on plastic bottle detection by Wang et al. (2018) as shown in Figure 3.6 can be attributed to three main factors. First, the use of a dataset collected from the local environment played a big role. This dataset included various examples that matched the real-life conditions, making the model to better understand and recognize future unseen plastic bottles.
In addition, using the YoloV5 algorithm in this study proved to be crucial. YoloV5 is known for its high performance in detecting objects. With a confidence threshold of 0.5, the model achieved an impressive accuracy of 99.5%. By combining a locally gathered dataset, accurate annotations, and the power of YoloV5, the model improved significantly compared to other models like YOLOv2, SSD, and Faster R-CNN (Wang et al., 2018). This shows that collecting specific and accurate data, along with choosing the right algorithm, can greatly enhance the accuracy of object detection models.
3.10.2 Introduction of Weight as Differentiating Factor of Plastic Bottle from Glass Bottle
In addition to improving the model's accuracy, this project stood out from similar studies that focused only on visual features for detecting plastic bottles. Unlike the studies by Dhulekar et al. (2018) and Wang et al. (2018), which relied solely on computer vision, this project introduced the weight of the bottles as an additional factor. This is important because plastic bottles and glass bottles have similar shapes, making it challenging to distinguish between them using visual features alone.
3.10.3 The Triplet Waste Bin Uniqueness in Relation to Recently Launched Similar Projects in the University of Ibadan
The completed ‘triplet waste bin” of this study as shown in Plate 3.1 focused on collecting three types of wastes: biodegradable wastes, plastic bottle wastes, and non-plastic bottle non-biodegradable wastes. Unlike recent projects like the Green Campus Project and The Junkyard Project, which aimed at ensuring the complete collection and recycling of all bottles and cans in the University of Ibadan community, this study goes beyond plastic bottles and cans. It considers the diverse range of wastes that people may come with.
The presence of triplet waste bins at every location within the University of Ibadan plays a crucial role in facilitating proper waste disposal and preventing mixing of different waste types. This study proffers a system that enables individuals to dispose wastes appropriately without muddling them up. The collected wastes in each bin can then be effectively sorted, promoting recycling initiatives and contributing to a more sustainable environment for all.