Paper Title
INTEGRATION OF CNN MODEL FOR IMAGE DETECTION AND MANIPULATION USING DOBOT MAGICIAN

Abstract
Abstract - The combination of robots and techniques of machine learning has led to novel prospects across a number of industries. The study emphasises on integration of Dobot Magician, a highly précised robotic arm with CNN model for image detection and manipulation. To enable the Dobot to draw, it needs to identify the image by enriching the features of the object. To make object detection simpler, grey scaling is utilised. Binarization and thresholding are then used to create contours, and ML methods like CNNs are used to categorise object characteristics. A labelled dataset is used to train the CNN model to divide the data into categories of cat, hen, horse, and elephant. In the proposed work, the effects of changing the hyperparameters under consideration and the outcomes they produce are compared. Optimizer function Adam is being used instead of RMSProp, batch size is set to 42, epochs are extended to 120, and the dataset is made more varied by photographing animals from various perspectives. The aforementioned adjustments allow the model to categorise a given picture with accuracy, precision, and recall of 92.90%, 92.71%, and 92.01%, respectively. After undergoing classification, the test image passes through pre-processing step, resulting in the generation of a contour map. This contour map is subsequently utilized by the dobot's drawing module. The contour map is used by dobot to create a drawing that exhibits similarities to the original input image. Keywords - CNN, Gray-Scaling, Binarization, Dobot, Contour Generation