Open AccessAn Automated Mammals Detection Based on SSD-Mobile Net
Author(s) -
Elham Mohammed Thabit A. Alsaadi,
Nidhal K. El Abbadi
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1879/2/022086
Subject(s) - computer science , artificial intelligence , object detection , pattern recognition (psychology) , computer vision , convolutional neural network , object (grammar) , digital image , process (computing) , similarity (geometry) , image (mathematics) , convolution (computer science) , minimum bounding box , field (mathematics) , image processing , artificial neural network , mathematics , pure mathematics , operating system
Animal detection techniques are one of the researcher’s interests and challenge. There are many difficulties faces by the researchers in this field that reduce the detection performance and efficiency, such as variation of image illumination, animal occlusion, the similarity of animal colors with background environment, etc. Multi-label Image Detection and classification of Mammals animals is the goal of this paper which we proposed to achieve in this proposal by using Single Shot Multi-Box Detector (SSD) and MobileNet v1 coco_2017 model. Localizing and classifying multiple objects (animals) of the Mammal category in digital images is another goal. The suggested SSD is regarded as a more accurate, fast, and efficient way to detect objects of different sizes based on deep learning technology. In this proposal, we used 2000 images in the network were collected from the standard dataset (such as Caltech 101) and the net. The SSD framework improves the detection and recognition processes of Convolution Neural Network (CNN). During the prediction time, the network introduces scores to the presence of every object class and bounded each object in the image with a box. Each box has a label that indicates the type of the object and the score represents the probability of the relationship of the object to that type. Boxes during the process are modified for getting the best matching to the object’s shape. The experimental results of this work proved the efficiency of classifying and detecting animals even in the variation of illumination, pose, and occlusion. Detection and classification accuracy is up to 98.7 %. This suggestion is more reliable and accurate than other similar works and detects a wide range of Mammals animals, unlike other similar works.