| At present,many old residential communities in China still use traditional mechanical water meters with wheels to monitor water consumption,which requires professional meter readers to come to their homes for meter reading and billing.Although meter readers use handheld meter reading terminals to read meters and bill,which avoids the occurrence of "human affairs meters",there are still problems such as low meter reading efficiency,heavy workload and high costs.At the same time,although the scope of application of intelligent IOT water meter is expanding,but due to the high cost of meter transformation and replacement,long replacement cycle and the difficulty of factors such as water companies in the comprehensive replacement of products,encountered greater resistance.In summary,to address the current problem that traditional character-wheel mechanical water meters cannot realize remote automatic meter reading,this paper proposes a deep learning-based side-side meter reading system scheme without modifying the meter body of character-wheel mechanical water meters.The scheme acquires the water meter display area through the camera of the smart meter reading terminal,and then uses a lightweight convolutional neural network model to recognize the digital characters before uploading the water meter values to the cloud server through NB-Io T communication technology to realize data storage and analysis.The main work of this paper includes the following aspects:(1)With Kendryte K210 artificial intelligence chip as the core processor and NBIo T communication technology for data transmission with the cloud server,the hardware platform of the intelligent meter reading terminal is designed and implemented,and software is developed for image pre-processing,network model inference and terminal data reporting.Meanwhile,an intelligent meter reading terminal structure is designed,which can place all the hardware devices of this system and also provide a stable light source to reduce the interference of environmental factors on the performance of the network model.(2)A self-made water meter digital character dataset containing full digital characters and double-half digital characters,and a digital recognition model based on SE-Mobile Net network are proposed.The network model uses Mobile Net lightweight network as the backbone network,and performs optimization operations such as improving the network structure and adding channel attention module to reduce the number of parameters and computation of the network and improve the recognition rate of digital characters by the network model.(3)The SE-Mobile Net network model is filter pruned to delete the convolutional kernel with weak feature extraction ability.Second,the model parameters of the pruned network model are quantized,and the parameters and activation values in the convolutional layer are quantized from 32-bit single-precision floating-point fixed-point to 8-bit fixed-point,which further improves the inference speed of the network model and thus reduces the overall energy consumption of the system.(4)The functions of device access and management,data storage and visualization of smart meter reading terminals are realized based on Ali cloud Io T platform.Meanwhile,the system is tested to verify the system performance and the feasibility of the scheme in this paper.The experimental results show that the system has the advantages of low power consumption,stable communication and high recognition rate of digital characters. |
