For verification of this paper, please visit on http://www.socialresearchfoundation.com/innovation.php#8

This research aimed to create a Machine Learning model to classify waste as either organic or recyclable based on image data. It uses an artificial neural network and a convolutionalneural network to achieve this. The final ANN model has an accuracy of 86% while the CNN model has an accuracy of 83%.

The study aims to create a Machine Learning model that accurately recognises organic and recyclable waste using images, and helps in waste segregation by making it more accurate and efficient.

Using AI to help with waste management has been a point of discussion for many years. There have been many researches who have worked on models to detect or classify waste in order to help with waste management. Here are some prominent research papers on the topic :-

EfficientDet-D2 and EfficientNet-B2 - The researchers for this paper have created a model to localize litter, and then another one to classify it as one of bio, glass, metal and plastic, non-recyclable, other, paper, or unknown. The model has achieved an accuracy of 70% on waste detection and 75% on its test Dataset.

Olugboja Adedeji and Zenghui Wang’s model - This model uses the ResNet- 50 Convolutional Neural Network model and Support Vector Machine (SVM) which is used to classify the waste into different groups. The model has achieved an impressive accuracy of 87% being tested on a trash image dataset.

Review of Research on AI applications in Waste Management - This research paper is a review of 85 studies published between 2004 and 2019 about the use of AI in Waste Management. The paper provides analysis of many of the AI models that have been used in waste management, and also discusses the challenges and insights of applying AI techniques in waste management. Other than researchers, there have also been a lot of tech companies have found innovative ways to help segregate and manage waste with the help of machines. The best example of a company like this is Greyparrot, which is a tech company whose AI automatically identifies different types of waste, providing composition information and analytics to help facilities increase recycling rates. 

The data was collected from a dataset on Kaggle containing 22,500 images of organic and recyclable objects. These images were used to train the model to classify waste as either organic or recyclable based on its image. Hence, this is an image recognition project.

The dataset was split into 80% testing and 20% training. The ANN network had 4 dense layers, and the CNN network had 6 dense layers and 8 convolutional layers. The models contain the activation functions relu, selu, and softmax. For ANN, regularization is used for its construction to prevent overfitting.

Two types of Machine Learning models are used:

Artificial Neural Network

The study of complexity is achieved by ANNs’ capacity to replicate the dynamic interaction of several components at once. ANNs also have the ability to make individual judgments

rather than generalisations. Compared to traditional statistical procedures, these tools may have some distinct advantages. When properly chosen and applied, the family of ANNs enables the maximising of what can be inferred from the data at hand as well as from complex, dynamic, and multidimensional occurrences, which are frequently difficult to forecast using the conventional ”cause and effect” concept.

Artificial neurons constitute ANNs. Each artificial neuron in the figure has a processing node, along with connections to other neurons, or ”dendrites” and ”axons,” from other artificial neurons. The multilayer perceptron, a popular ANN architecture, organises the neurons in layers. The input layer is given an ordered set (a vector) of predictor variables.Each input layer neuron shares its value with every neuron in the middle layer. There is a connection weight along each connection between the input and middle neurons, and the middle neuron receives the product of the value from the input neuron and the connection weight.

Each intermediate layer neuron adds up its weighted inputs before applying a non-linear (often logistic) function. The output from that specific middle neuron is the result of the function. The output neuron is connected to each middle neuron. There is a connection weight along each link between a middle neuron and the output neuron. The output neuron then applies the non-linear function to the weighted sum of its inputs as the last step. The output of the entire ANN is the outcome of this function.

Convolutional Neural Network

Convolutional Neural Network (CNN) architecture is one of the most impressive types of ANN architecture. CNNs are primarily used to tackle complex image-driven pattern recognition problems and, thanks to their accurate yet simplistic architecture, provide a streamlined way to get started with ANNs.

The overall architecture of the Convolutional Neural Network (CNN) includes an input layer, multiple alternating convolution and max-pooling layers, one fully-connected layer and one classification layer. The traditional CNN structure is mainly composed of convolution layers, pooling layers, fully connected layers, and some activation functions. Each convolution kernel is connected to the part of feature maps. The input is connected to all of the output elements in the fully connected layer.

Multiple convolutional networks will be blended with nonlinear and pooling layers in the network. The output of the first convolution layer becomes the input for the second layer after the image has gone through one convolution layer. Additionally, this occurs with each additional convolutional layer. After each convolution operation, the nonlinear layer is introduced. It has a function called activation that adds nonlinear properties. Without this characteristic, a network would not be strong enough to model the response variable (as a class label).

The nonlinear layer is followed by the pooling layer. The image’s width and height are used as input, and a downsampling process is carried out on them. The image

volume is consequently decreased. As a result, if certain features (such as boundaries, for example), have already been recognised in the previous convolution operation, a detailed image is no longer required for further processing and is compressed into less detailed images.

It is important to attach a fully connected layer once a series of convolutional, nonlinear, and pooling layers have been finished. Convolutional networks’ output data are used in this layer. An N-dimensional vector, where N is the number of classes the model was built from, is produced by adding a fully linked layer to the network’s end. The number of classes from which the model chooses the desired class, N, is produced when a completely linked layer is attached to the network’s end.

The results were somewhat surprising. The ANN network achieved an accuracy of 86% while the CNN network, which is supposed to perform better, achieved an accuracy of only 83%. This indicates that perhaps there may have been overfitting in the CNN network, something to improve upon in the model. The problem of effective waste management is a plaguing issue of our times. As an increasing number of professions move toward automation, waste management should also be a part of the movement. The model that has been created is satisfactorily accurate, but still has room to be worked upon. The convolutional neural network needs to be tuned more to prevent overfitting. The artificial neural network, is a promising model to implement in the field of waste management. The researchers hope that this paper incites further research in this field that is helpful to the technological advancement in waste management.

1. Grossi, Enzo Buscema, Massimo. (2008). Introduction to artificial neural networks. European journal of gastroenterology hepatology. 19. 1046-54. 10.1097/MEG.0b013e3282f198a0. 2. Tangri, Navdeep Ansell, David Naimark, David. (2008). Predicting technique survival in peritoneal dialysis patients: Comparing artificial neural networks and logistic regression. Nephrology, dialysis, transplantation : official publication of the European Dialysis and Transplant Association - European Renal Association. 23. 2972-81. 10.1093/ndt/gfn187. 3. Raschka, Sebastian. (2019). BogoToBogo. https://www.bogotobogo.com/python/scikit-learn/Artificial-Neural-Network-ANN-7-Overfitting-Regularization.php 4. O’Shea, Keiron Nash, Ryan. (2015). An Introduction to Convolutional Neural Networks. ArXiv e-prints. 5. A State-of-the-Art Survey on Deep Learning Theory and Architectures - Scientific Figure on ResearchGate. Available from: https://www.researchgate.net/figure/The-overall-architecture-of-the-Convolutional-Neural-Network-CNN-includes-an-input_fig4_331540139. 6 Kang, Xu Song, Bin Sun, Fengyao. (2019). A Deep Similarity Metric Method Based on Incomplete Data for Traffic Anomaly Detection in IoT. Applied Sciences. 9. 135. 10.3390/app9010135. 7. Sorokina, K. (2019, February 26). Image classification with Convolutional Neural Networks. Medium. Retrieved July 22, 2022, from https://medium.com/@ksusorokina/image-classification-with-convolutional-neural-networks-496815db12a8 . 8. Choubey, V. (2020, July 22). Text classification using CNN. Medium. Retrieved July 22, 2022, from https://medium.com/voice-tech-podcast/text-classification-using-cnn-9ade8155dfb9. 9. Yann, L., Bengio, Y., Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444. https://s3.us-east-2.amazonaws.com/hkg-website-assets/static/pages/files/DeepLearning.pdf. 10 . Sashaank Sekar. (2019). Waste Classification Dataset. https://www.kaggle.com/datasets/techsash/waste-classification-data