| With the rapid development of China’s economy,consumers are demanding higher and higher quality of fresh products.The quality assurance of food products transported over long distances relies heavily on storage conditions during transportation,leading to the growth of the cold chain logistics industry.Cold chain logistics is a kind of logistics industry,unlike ordinary logistics,it uses cold chain containers to transport temperaturesensitive products such as fruits and vegetables,aquatic products,frozen products and certain drugs.To ensure that the cold chain container works properly,the internal temperature of the container needs to be collected and monitored.At present,the manual inspection method is adopted in the cold chain container transit terminal,it requires the practitioners to go to read the temperature information on the container dashboard to ensure the working status of the container refrigeration equipment at a fixed period of time.It is labor-intensive that relying on manual inspection method to read and record the temperature information on the dashboard.Therefore,there is an urgent need for an automated collection scheme to replace the manual inspection scheme and reduce the work intensity and difficulty of practitioners.In response to this problem,many researchers have proposed an Io T-based temperature monitoring scheme,in which temperature collection nodes are placed in cold chain containers to collect internal temperature information in real time,and upload temperature information to a cloud server through Io T technology to achieve monitoring of container temperature.This scheme poses a challenge to communication reliability,and such many nodes are not easy to manage and maintain.Therefore,the practical value of the system is not guaranteed.An image-recognition-based temperature monitoring scheme is designed and implemented after investigating the actual harbor terminal container transshipment sites in this thesis.The scheme takes pictures of containers by moving the orbital robot,identifies the temperature information on the dashboard using an image recognition algorithm,and uploads it to the server for the management staff to view.The main work of the thesis is as following:(1)Design and construct a track-mounted robot equipped with lateral and longitudinal motors for controlling movement,a camera for capturing dashboard images,a laser obstacle avoidance module for collision avoidance,a magnetic induction sensor for position information,and a power management module for real-time power monitoring.The Raspberry Pi 4b is used as the main control module to realize robot control and temperature recognition functionalities,etc.(2)The data set required for the target detection algorithm is produced,with 1146 pictures of the dashboard and 954 pictures of the temperature figures labeled.(3)A temperature recognition method based on two-step target detection is proposed.The first step is to detect the category and location of the dashboard and crop down the dashboard area;the second step is to detect the digits on the cropped area and select different combination algorithms by dashboard type to combine the digits into temperature,which solves the problems of different dashboard temperature arrangements and the difficulty of detecting small digit targets.(4)The target detection network YOLOv4-tiny is optimized,and three sizes of backbone networks such as original network size(L),digital detection network size(M),and dashboard network size(S)are designed to adapt to the demands of tasks with different recognition difficulties,which improves the inference speed of the model without reducing the average accuracy.Furthermore,attention mechanism is added to the network to improve the average accuracy. |
