CN211699025U - Novel artificial intelligence edge computing equipment based on FPGA - Google Patents

Abstract

The utility model discloses a novel artificial intelligence edge computing equipment based on FPGA, including convolution neural network, convolution neural network is connected with FPGA hardware automatic transmission electricity, the FPGA hardware includes FPGA computational element, be connected with the DDR buffer memory on the FPGA computational element, the last transmission electricity respectively of FPGA computational element is connected with standard HDMI input and standard HDMI output, standard HDMI input is connected with input device transmission electricity, standard HDMI output is connected with display transmission electricity, standard HDMI input and standard HDMI output are connected with distal end server transmission electricity through the delivery network respectively; by using the method and the system, a user can independently call the basic computing library according to the characteristics of the algorithm to flexibly realize different algorithms, and the problem that non-FPGA developers spend a large amount of time to deploy the algorithm to a hardware system is avoided, so that the system development time is saved.

Description

Novel artificial intelligence edge computing equipment based on FPGA

Technical Field

The utility model relates to an artificial intelligence technical field specifically is a novel artificial intelligence edge computing equipment based on FPGA.

Background

Most of the current artificial intelligence computing hardware systems are mostly composed of a Graphic Processing Unit (GPU) or a GPU system, and comprise a display, a PC, a high-performance CPU or GPU and the like. And interconnection is carried out through a network, so that real-time input and analysis of the original data are realized.

With the rapid growth of internet users and the rapid expansion of the data volume, the computing pressure of data centers is increasing. Meanwhile, the rise of computation-intensive fields such as artificial intelligence, high-performance data analysis, financial analysis and the like has greatly exceeded the capability of the traditional CPU processor, edge computation based on the FPGA is developing rapidly in order to share the computation pressure of a data center, the edge computation needs very high computation performance, and has very high real-time performance, the FPGA can process a lot of things simultaneously and in parallel, the requirement of data processing speed can be met, and the requirement of real-time performance can also be met.

In the aspects of face detection and road condition vehicle detection, most of the existing equipment adopts a design scheme of a CPU (Central processing Unit) and a GPU (graphics processing Unit), a PC (personal computer) or a workstation needs to be configured, and the system is large in power consumption and large in size and is difficult to flexibly deploy. And the application system adopting the FPGA can realize very low power consumption in a small-volume mode and meet the requirement of real-time property.

The FPGA has flexible programmable characteristic and can be adapted to various deep learning algorithms, but the transplantation of various deep learning algorithms on an FPGA platform has high requirements on the development skills of the FPGA and an embedded operating system of developers, the debugging time is long, and the debugging process is complicated. Therefore, the method improves the method and provides novel artificial intelligence edge computing equipment based on the FPGA.

SUMMERY OF THE UTILITY MODEL

For solving the defect that prior art exists, the utility model provides a novel artificial intelligence edge computing equipment based on FPGA.

In order to solve the technical problem, the utility model provides a following technical scheme:

the utility model relates to a novel artificial intelligence edge computing equipment based on FPGA, including convolution neural network, convolution neural network is connected with FPGA hardware automatic transmission electricity, the FPGA hardware includes FPGA computational element, be connected with the DDR buffer memory on the FPGA computational element, last difference transmission electricity of FPGA computational element is connected with standard HDMI input and standard HDMI output, standard HDMI input is connected with input device transmission electricity, standard HDMI output is connected with display transmission electricity, standard HDMI input and standard HDMI output are connected with distal end server transmission electricity through carrying the net respectively.

As a preferred technical solution of the present invention, the input device includes a player and a camera.

As an optimal technical solution of the present invention, the transmission network includes light transmission and ethernet transmission.

As an optimized technical scheme of the utility model, FPGA includes deep learning basic calculation unit and system control unit.

As an optimal technical scheme of the utility model, the convolution neural network passes through automatic script, and the convolution neural network file under the automatic identification caffe frame converts it into the inside calculation module of FPGA.

The utility model has the advantages that: according to the novel artificial intelligence edge computing device based on the FPGA, on the basis that the FPGA is used as an edge computing core device, a basic computing base of a deep learning algorithm is designed and integrated into the FPGA device, a user can automatically call the basic computing base according to the characteristics of the algorithm to flexibly realize different algorithms, the situation that non-FPGA developers spend a large amount of time to deploy the algorithm to a hardware system is avoided, and therefore system development time is saved.

Drawings

The accompanying drawings are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention, and together with the description serve to explain the invention and not to limit the invention. In the drawings:

fig. 1 is a block diagram of a novel artificial intelligence edge computing device based on FPGA of the present invention;

fig. 2 is a schematic diagram of an image transmission information working square of the novel artificial intelligence edge computing device based on the FPGA of the present invention;

FIG. 3 is a schematic diagram of an image display of a novel FPGA-based artificial intelligence edge computing device of the present invention;

FIG. 4 is a schematic diagram of the rapid deployment of the deep learning algorithm of a novel FPGA-based artificial intelligence edge computing device of the present invention;

FIG. 5 is a schematic diagram of the FPGA signal portion of a novel FPGA-based artificial intelligence edge computing device of the present invention;

FIG. 6 is a schematic diagram of a FPGA power supply portion of a novel FPGA-based artificial intelligence edge computing device of the present invention;

fig. 7 is a schematic circuit diagram of a DDR cache portion of the novel artificial intelligence edge computing device based on FPGA of the present invention;

FIG. 8 is a schematic diagram of a gigabit Ethernet portion of a novel FPGA-based artificial intelligence edge computing device of the present invention;

fig. 9 is a schematic circuit diagram of the video output part of the novel artificial intelligence edge computing device based on FPGA.

Fig. 10 is a schematic diagram of a gigabit fiber input part of the novel artificial intelligence edge computing device based on FPGA of the present invention.

Detailed Description

The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are presented herein only to illustrate and explain the present invention, and not to limit the present invention.

Example (b): as shown in fig. 1-10, the utility model relates to a novel artificial intelligence edge computing device based on FPGA, including convolution neural network, convolution neural network is connected with FPGA hardware automatic transmission electricity, the FPGA hardware includes FPGA computational element, be connected with the DDR buffer memory on the FPGA computational element, it is connected with standard HDMI input and standard HDMI output to transmit electricity respectively on the FPGA computational element, standard HDMI input is connected with input device transmission electricity, standard HDMI output is connected with display transmission electricity, standard HDMI input and standard HDMI output are connected with distal end server transmission electricity through carrying the net respectively.

The input device comprises a player and a camera.

Wherein, the transport network comprises optical transmission and Ethernet transmission.

The FPGA comprises a deep learning basic computing unit and a system control unit.

The convolutional neural network automatically identifies convolutional neural network files under a mask framework through an automatic script and converts the convolutional neural network files into a calculation module inside the FPGA.

The working principle is as follows:

the automatic deployment of the convolutional neural network to FPGA hardware can be realized, the detection and display of objects are realized, and the detection and display results can be backed up to a remote server or shared to other edge computing equipment.

The standard HDMI input, the standard HDMI output, the DDR cache, the FPGA computing unit, the gigabit Ethernet interface and the optical fiber interface are utilized to realize the input and the output of standard HDMI interface equipment and realize the automatic target search in real-time monitoring.

And automatically deploying the convolutional neural network to FPGA hardware, automatically identifying the convolutional neural network file under a mask framework through an automatic script, and converting the convolutional neural network file into a calculation module inside the FPGA.

The detection and display of the object can realize the detection of vehicles, pedestrians, airplanes and human faces, and the detection is displayed through the display.

And backing up the detection and display results to a remote server or sharing the detection and display results to other edge computing equipment, wherein the detection and display results are interconnected with other edge computing equipment in an Ethernet networking mode, and are interconnected with the remote server through a gigabit Ethernet.

The target automatic search in real-time monitoring is realized, the far-end image acquisition information can be accessed through the optical fiber, the near-end image acquisition information can be accessed through the gigabit Ethernet, real-time processing is performed through a DDR cache and an FPGA internal algorithm, and a target object to be monitored and found is displayed to a display end through an HDMI interface in real time.

The method comprises the steps that video stream data are input in real time through a standard HDMI cable, an HDMI interface can be externally connected with a player or a camera, and the video stream enters equipment, then is cached through a DDR (double data rate) and enters an FPGA (field programmable gate array) for calculation; the FPGA is internally integrated with a deep learning basic calculation unit, and the calculation result of the video stream data is superposed to the original video stream image through an algorithm identifier, is sent to a display through a standard HDMI interface, and can be sent to a remote service through a gigabit Ethernet or a light interface.

In the process of deploying the deep learning algorithm to the FPAG hardware part, the equipment provides an algorithm conversion tool based on a cafe frame, and can convert a basic convolution neural network unit into a basic operation unit in the FPGA, so that the aim of rapidly deploying the artificial intelligent deep learning algorithm on the FPGA is fulfilled.

In a hardware system, a video stream is converted into original data through a PHY chip of an HDMI and is cached in a DDR memory of equipment, a system control unit of an FPGA calls a cache image in the DDR memory in real time and sends the original image to a deep learning basic operation unit, the basic operation unit automatically combines according to a deep learning algorithm and calculates video image data to obtain a calculation result, the calculation result can be singly or superposed to the original image and cached in the DDR memory, and the system control unit sends the processed image to a display through an HDMI interface for displaying.

As shown in fig. 3, a video image is input into a device, a video stream is decoded into raw data in RGB format by a PHY chip of HDMI in a hardware device, the raw RGB data is cached in a DDR memory by a video image cache, the RGB data cached in the DDR memory is sent to a basic operation unit by a control unit of an FPGA, a calculation result is superimposed on an image in RGB format and sent to the DDR memory for caching, and the cache result is converted into a standard HDMI code stream by the PHY chip of HDMI and sent to a display for display.

As shown in fig. 4, in order to implement fast deployment of the deep learning algorithm, a deep learning model based on a cafe framework is automatically identified, convolution layers, pooling layers, full-link layers, BN layers, and activation function layer structures in the model are extracted, and according to parameter settings in the model, a basic calculation unit inside the FPGA is automatically called and mapped into a hardware design of the FPGA.

As shown in fig. 2, in order to achieve compatibility between long-range and short-range image transmission information, two image input modes, namely an optical fiber and a gigabit ethernet, are designed, where the optical fiber completes image data input to the far-end image acquisition device, and the gigabit ethernet completes image data input to the near-end image acquisition device. And after entering the equipment, the image data is cached through the DDR, real-time algorithm processing is carried out inside the FPGA, a target to be detected is marked on the original image in real time, and the target is displayed on an external display through the HDMI.

Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described in the foregoing embodiments, or equivalents may be substituted for elements thereof. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (4)

1. The utility model provides a novel artificial intelligence edge computing equipment based on FPGA, its characterized in that, includes the convolution neural network, the convolution neural network is connected with FPGA hardware automatic transmission electricity, the FPGA hardware includes FPGA computational element, be connected with the DDR buffer memory on the FPGA computational element, on the FPGA computational element respectively transmission electricity be connected with standard HDMI input and standard HDMI output, standard HDMI input is connected with input device transmission electricity, standard HDMI output is connected with display transmission electricity, standard HDMI input and standard HDMI output are connected with distal end server transmission electricity through carrying the net respectively.

2. The FPGA-based novel artificial intelligence edge computing device of claim 1, wherein the input device comprises a player and a camera.

3. The FPGA-based novel artificial intelligence edge computing device of claim 1, wherein the transport network comprises optical and Ethernet transport.

4. The FPGA-based novel artificial intelligence edge computing device of claim 1, wherein the FPGA comprises a deep learning basic computing unit and a system control unit.

Images