An empirical study and data analytics using deep learning techniques in garbage classification

Abstract

A profound transformation in waste composition arises from significant newlineshifts in lifestyle, spurred by intense competition among consumer goods newlineproducers and innovative strides in packaging technology. Overflowing bins newlinecontaminate soil, leaving the extent of waste unaccounted for within cities. newlinePrecisely identifying waste types remains challenging; current predictive image newlineclassification lags, leading to prolonged identification processes. To address newlinethis, a modified ResNeXt model, featuring a quothorizontal and vertical blockquot newlineaddition to expand network capacity without significantly inflating parameters, newlineis employed. Branches with diverse convolutional layers capture a broader array newlineof input image features, enhancing classification performance. Additional dense newlineand dropout layers are integrated to refine the standard ResNeXt model. Model newlinepruning optimizes network connections and reduces overall size. The newlinearchitecture, trained on garbage images, concurrently deploys convolution newlineneural networks modified ResNeXt for biodegradable, non-biodegradable, newlineand hazardous waste, and ResNet 50 for other waste types. The system attains newline98% accuracy in rapid garbage identification, surpassing existing deep learning newlinemodels. The proposed three-bin smart system, equipped with sensors, motors, newlineand Raspberry Pi technology, segregates waste efficiently, aids in waste newlinecollection, and sustains environmental cleanliness. By predicting and newlinecategorizing waste generation, the model achieves 98.9% accuracy, newlineoutperforming current methods. The proposed innovative approach minimizes newlinemanual labor, optimizes waste disposal, and promotes environmental health. newline