An Ensembling Approach for Efficient Waste Classification

One of the major challenges faced by the recycling industry is waste segregation. Unsegregated wastes are not favorable for the environment and manual segregation is quite harmful to the health of the engaged workforce. Therefore, this paper aims to propose an efficient waste classification model to classify and identify the different types of waste. Convolution Neural Network-based models such as VGG16, MobileNetV2, In-ceptionV3, DenseNet201, and ResNet152V2, trained on ImageNet have been considered for the weighted average-based ensembling technique to classify waste images. Five approaches based on accuracy, specificity, precision, recall, and F1-score are used to calculate the weight of each model to evaluate the performance metrics of the proposed model. The F1-based approach for weight calculation of the models outperforms the other existing CNN models by achieving an average performance of 93.881%.