Research And Implementation Of Intelligent Supervision Function Of Android Platform Based On TensorFlow

With the continuous development of deep learning and artificial intelligence,there are more and more related applications in all walks of life.At present,the traditional supervision work is gradually becoming intelligent,and inspection systems using deep learning algorithms have been developed one after another,and can automatically supervise inspection personnel and equipment.However,existing systems often have the following deficiencies.Due to the limitation of its own computing and storage capabilities,the mobile terminal generally runs on the computer.The computer is easily restricted by the location and usage scenarios,which is inconvenient to use.When the OCR model performs text detection on the image of the instrument and instrument in the inspection work,it is easy to use The status light on the device is incorrectly detected as text;the YOLO algorithm presets that the Anchor box needs to randomly select k center points,which is affected by human factors.In view of the above shortcomings,this paper mainly conducts in-depth research on the text detection related algorithms in the OCR model,and determines the OCR model in this paper in conjunction with the multi-target detection algorithm;in-depth research on the Anchor mechanism in the existing YOLO algorithm,combined with-Means++aggregation algorithm and IOU distance were preset size and size of Anchor box;designed and implemented Android platform intelligent supervision system based on TensorFlow.This article mainly completed the following parts:(1)The key technologies used in intelligent supervision include OCR model and target detection technology.By analyzing the text detection process in the OCR model and the characteristics of the target in the image of this article,the multi-target detection algorithm is used to replace the original text detection algorithm in the OCR model to determine the OCR model in this article.The model detects text while detecting other information in the image of this article,which simplifies the existing system process of detecting multiple targets.The existing target detection algorithm is also analyzed,and the target detection algorithm YOLO used in this paper is determined through comparative analysis.(2)In order to pre-set the size and size of the Anchor box in the YOLO algorithm due to the influence of human factors,the K-Means++algorithm and the IOU distance are used for presetting.After optimization,the effect of randomly selecting k initial values can be reduced,and the convergence of the algorithm can be accelerated to achieve the desired effect.(3)According to the characteristics of text and other target recognition in the inspection work,the Android platform intelligent supervision system based on TensorFlow is designed and implemented.This module mainly includes four parts:image acquisition,preprocessing,information recognition and display,and has been tested systematically to prove the practicability and accuracy of the system.