Garbage Separation Using Object Detection开题报告

1. 研究目的与意义（文献综述包含参考文献）

Organic waste is a type of waste that is easily biodegradable. We have to separate organic waste from inorganic waste. Organic waste can be buried in the soil as fertilizer, while inorganic waste can be recycled. The decomposition of organic waste will occur faster than inorganic waste (widiana 2017)The author conducted an interview with Mrs. Indah as the head of a garbage bank in Pamulang, South Tangerang. Ibu Indah said, there are still many residents who don't care about the condition of the garbage. To provide a solution to this problem, the researchers obtained data from Mrs. Indah as the head of the garbage bank in Pamulang, South Tangerang. which states the importance of making an accurate waste sorting tool in sorting organic and non-organic waste. To be able to make this tool, machine learning assistance is needed, where the use of machine learning is to help process further classified data. The training data consists of a series of examples, where each example is represented as a vector pair of input (feature) and the desired output value (Parmar, Grossmann, Bussink, Lambin, Aerts, 2015). Support Vector Machines (SVM) are used as a related learning algorithm that analyzes the data used for classification and regression analysis. By providing data sets for training. (Cortes Vapnik, 1995).From the results of the analysis, literature study, interviews, the background of the existing problems is that the researcher makes the idea of a system or tool that can automatically sort organic and non-organic waste with the aim of facilitating the work of the environmental service in processing waste around the environment. Therefore, the authors conducted a research entitled "Garbage Separation Using Object Detection"

2. 研究的基本内容、问题解决措施及方案

Formulation of the problem Based on this background, the problems in this thesis can be formulated, namely: a.How to make a robot design for sorting organic and non-organic waste automatically? b.How do robots work in sorting organic and inorganic waste? c. What is the level of success of the robot in terms of accuracy that distinguishes organic and non-organic waste? Research methodology The method used by the author in this study is divided into two, namely the method of data collection and method of development. The following is an explanation of the two methods: Method of collecting data In conducting data analysis and writing this thesis, the authors use 2 data collection methods, namely: 1.Library study by collecting theories, concepts, and information from books, journals, the internet, and similar literature. 2.Field study with observation, interviews System Development Methods In supporting research, the material used for research is data sample image of garbagetook from outside as a training sample (training) and the results of testing (testing) in the form of real-time video. The initial stage conducted in this study using 2 (two) platforms by separating the training environment (training) and the testing environment (testing). Training using a laptop device with a python programming language and based CLI (command line interface). Any convolutional operation, apart from has a filter kernel size parameter and the number of filters, it also has a parameter others that affect the shape of the output resulting from the convolutional operation, namely padding and stride. Padding is a parameter that represents an amount added a border around the edges of the input, which minimizes the function loss of information at the edges of the image (input). This is due to the convolution process itself, where the edges of the image are usually passed by the filter kernel, except for the 1x1 kernel. The popular and simplest method for solving this problem is the use of zero-padding, that is, adding values 0 on each edge of the image (input). Whereas stride is a parameter represents the number of shifts (steps) performed by the filter kernel. This parameter is commonly used to reduce the size of the output. YOLO is a new approach to object detection systems, that is targeted for real-time processing. YOLO frames detection object as a single regression problem, where from the image pixels directly to a separate spatial bounding box and a probability class related. YOLO performs object detection and recognition with a single neural network (single neural network), which predicts the boxes constraints and class probabilities directly in one evaluation (Redmon etal., 2015). To get the final prediction, the determining factor is class The confidence score is obtained, based on the conditional probability of classes and boxes confidence score. Class confidence score measures the value of confidence in classification and localization of objects. Class confidence score gives a confidence value specific class for each box, which encodes the possible classes that are appears in the box and how well the predicted box matches the object.