A novel and advanced deep learning model for classifying municipal solid waste images

ABSTRACT The issue of garbage disposal has been explored many folds in the previous years due to the massive disposable goods production in almost every organization. An effective handling of these wastes is one of the primary issues of advanced society. Traditional waste classification models highly depend on manual separation of objects that are tiresome, expensive, and prone to subjective errors created by poor expertise of waste categorization and inefficiency. However, the growth of artificial intelligence has resulted in the incorporation of machine learning strategies to increase waste classification accuracy. Nevertheless, the present models have poor ability to autonomously classify waste categories at the household level, resulting in unacceptable waste disposal. Moreover, conventional automated models highly depend on outdated strategies and lack recent classification mechanisms. While automatic solid waste classification models offer satisfactory solutions, complexities remain in their technological advancement and implementation, urging ongoing research and development. In this context, an effective multi-classification model of municipal solid waste using a deep learning approach is developed to increase the effectiveness and accuracy of municipal solid waste classification processing. First, the necessary images are collected from the benchmark dataset. After the image collection, the garnered images are transferred to an adaptive vision transformer-yolov5-based object detection model. This model leverages the strength of the vision transformer and Yolov5. Here, the vision transformer improves the feature extraction process for better representation of complex data and YoloV5 helps to obtain contextual information to increase the performance in complex scenes. To enhance the effectiveness of the object detection operation, the hyperparameters in the developed model are optimized by using the enhanced dung beetle optimizer. Then, from the detected object images, the region of interest is extracted. After this, the extracted images are given as input to the designed dilated residual attention network. The model is incorporated with a residual attention network and a dilated convolution mechanism. Here, the residual attention network aids in focusing the most relevant part obtained from the extracted image leading to improved accuracy in the classification process, and the dilated convolutional layers in this network have the ability to retrieve robust features for the classification process. Finally, the classified outcome and the performance of the developed model are validated with numerous traditional systems. The accuracy of the implemented municipal solid waste multi-classification process is 94%. Thus, the implemented municipal solid waste multi-classification process proved its economic sustainability and effectiveness over the traditional models.