Research On The Cattle Detection And Recognition Based On Android Device

Object detection is a very important task for different applications including autonomous driving,face detection,video surveillance,etc.Some of the latest object detection technologies can ensure both high recognition accuracy and high detection speed.They’re capable of localizing and identifying objects in real-time both in images and videos.But the most current deep learning applications are running on servers or desktop computers.The Android platform works not only to capture images,video through the camera,also able to detect animal faces.Considering there are a lot of mobile computing devices available,in this research is implemented the cattle face detection and recognition application based on object detection algorithm for Android devices.The current high performance of mobile devices can meet the computing requirements of deep learning,and making it possible to run target detection on Android devices.In this paper,the developed application will be able to detect and recognize cattle which will try to localize and identify not just one,multiple cattle face in the real-time.The model chooses for this project was optimized for speed,for the real-time portion of the code to work speed is very important.I have a training a pre-trained model called SSD-Mobile Net-V1 and currently running it via Tensor Flow.The Tensor Flow offers a pre-trained model for drawing bounding boxes around in detected images,together with a tracking code to follow them over time.The SSD-Mobile Net model can greatly reduce the number of parameters,and achieve higher accuracy.SSD network is a regression model,which uses features of different convolution layers to make classify regression and boundary box regression.The Mobile Net network was developed to improve the real-time performance of deep learning under limited hardware conditions.A combination of Mobile Net and SSD gives outstanding results in terms of accuracy and speed in cattle detection and recognition activities.The complete model contains four parts: the input layer for importing the target image,the Mobile Net base net for extracting image features,the SSD for classification regression and bounded box regression and the output layer for exporting the detection results.For my dataset,successfully trained a model using 7444 images of cattle and it is working with high accuracy.In this collected dataset includes 256 different cattle classes and about 40-50 pictures of each cattle.The dataset has been split into training and testing dataset.The 243000 training steps took roughly 4 days utilizing the Nvidia GTX 1060 in my computer system.After training,could see total loss get down to 0.410 and precision up to 0.95.The total confidence scores detection quality(total m AP)for all the cattle classes are 96-100% based on the SSD-Mobile Net-V1 model.The detection and recognition process is achieved using two methodsto evaluate the detection and recognition performance using Android camera(Galaxy S9)and the Tensor Flow object detection on the PC in terms of accuracy and detection speed.Experimental results have demonstrated valuable improvements in terms of detection accuracy and efficiency for cattle face identification.For reference,it’s using all cores at 80-100% of the Intel i5-4590 CPU @3.30 GHZ x4 in that system during the training process.