Research On Multi-band Signal Detection Technology And Application Based On Deep Learning

Object detection is an important research field in computer vision tasks,and it has been applied in many scenarios such as security,medicine and military.With the rapid development of urban intelligence,industrial intelligence,medical intelligence,etc.,urgent and strict requirements are placed on the speed and accuracy of object detection.In recent years,deep learning has continuously improved the detection accuracy of object detection tasks with its powerful autonomous feature learning ability.However,the traditional visible single-mode signal detection limits the application of object detection in complex scenes.At the same time,the rapid development of information technology makes the data representation form also exhibits polymorphism,multi-source and multi-descriptive characteristics.Multi-band data such as visible light images,infrared images,laser signals can achieve data complementation,and the information provided by images or signals in different bands is redundant,complementary,and cooperative.Multi-band object detection has greatly expanded the application scenarios of deep learning and has important research value.This paper proposes three improved deep learning algorithms for short-band one-dimensional medical signal,visible light band two-dimensional face images and infrared band urban scene images.Based on engineering applications and algorithm research,this paper establishes a multi-band object detection system and realized the actual scene application.The main research work of this paper is as follows:(1)A short-band EEG detection algorithm based on an improved convolutional neural network is proposed to realize quantitative automatic diagnosis of brain in attention deficit hyperactivity disorder(ADHD)patients.Due to the complexity and unpredictability of brain lesions,there are still some brain diseases that cannot be quantitatively diagnosed in clinical practice,which brings great obstacles for further diagnosis and treatment.Aiming at the limitation that patients with ADHD can only be diagnosed qualitatively,control people' and patients' clinical EEG data were collected,and an improved deep learning algorithm was proposed to detect one-dimensional lesion signals.Experiments show that the detection of short-band EEG signals in this paper can effectively detect patients with ADHD.The proposed method has high diagnosis rate,good quantification,and can effectively assist in the diagnosis of clinical brain diseases.(2)A visible light band object detection algorithm based on multi-task convolutional neural network(MTCNN)is proposed,which is applied to the face detection algorithm,and a software recognition system that can be applied to identity authentication and security protection is realized.Face detection is the most common and the most widely used in visible light band images.In this paper,a new face detection network is realized by proposing a MTCNN and combining FaceNet to fully extract the features of two-dimensional images in the visible band.Experiments show that this method has high detection rate and low false detection rate.This paper builds a set of face detection and recognition software system based on the algorithm,which has been applied to a variety of actual authentication scenarios and can meet the requirements of real-time face detection.At the same time,this paper also studies facial expression recognition.The designed VGG model completes the detection of facial expression,the extraction of feature points and the classification of facial expressions.In this paper,based on the face detection in the visible band and combined with deep learning algorithm,a series of face attribute recognition research tasks have been completed,which has high practical application value.(3)An infrared band object detection algorithm based on the improved YOLO(You Only Look Once,YOLO)algorithm is proposed,which is applied to pedestrian and vehicle detection,and real-time infrared object detection that can be applied to assisted driving and unmanned systems is realized.The brightness of the target in infrared imaging is almost unaffected by the light source and texture,and it is a necessary image acquisition method in night vision and insufficient visible light.This paper proposes an improved YOLOv3-Infrared algorithm.First,the style conversion network is used to convert the infrared image into a pseudo-visible light image to improve the image quality.Then,the infrared target is dimensionally clustered again.Next,the network pre-training process and the multi-scale training model are adjusted.Finally,the semantic enhancement branch of shallow feature maps is introduced to enhance the detection effect of low-resolution pixels and realize infrared object detection.The infrared object detection method proposed in this paper is applied to pedestrian and vehicle detection.Experiments show that the proposed method is significantly better than YOLOv3,and has good real-time performance,high accuracy and low missed detection rate.It can be used in assisted driving and unmanned systems to improve the safety of drivers and pedestrians traveling at night.In summary,three improved deep learning object detection methods for multi-band one-dimensional and two-dimensional signals are proposed,and the practical application of software systems are realized,which can be applied to assisted medical diagnosis and treatment,assisted automatic driving,security monitoring,and unmanned system,etc.This paper completed a series of theoretical research and engineering applications of object detection.