To address the problems of large storage requirements, computational pressure, untimely data supply of off-chip memory, and low computational efficiency during hardware deployment due to the large number of convolutional neural network (CNN) parameters, we developed an innovative hardware-friendly CNN pruning method called KRP, which prunes the convolutional kernel on a row scale. A new retraining method based on LR tracking was used to obtain a CNN model with both a high pruning rate and accuracy. Furthermore, we designed a high-performance convolutional computation module on the FPGA platform to help deploy KRP pruning models. The results of comparative experiments on CNNs such as VGG and ResNet showed that KRP has higher accuracy than most pruning methods. At the same time, the KRP method, together with the GSNQ quantization method developed in our previous study, forms a high-precision hardware-friendly network compression framework that can achieve “lossless” CNN compression with a 27× reduction in network model storage. The results of the comparative experiments on the FPGA showed that the KRP pruning method not only requires much less storage space, but also helps to reduce the on-chip hardware resource consumption by more than half and effectively improves the parallelism of the model in FPGAs with a strong hardware-friendly feature. This study provides more ideas for the application of CNNs in the field of edge computing.

To address the problems of convolutional neural networks (CNNs) consuming more hardware resources (such as DSPs and RAMs on FPGAs) and their accuracy, efficiency, and resources being difficult to balance, meaning they cannot meet the requirements of industrial applications, we proposed an innovative low-bit power-of-two quantization method: the global sign-based network quantization (GSNQ). This method involves designing different quantization ranges according to the sign of the weights, which can provide a larger quantization-value range. Combined with the fine-grained and multi-scale global retraining method proposed in this paper, the accuracy loss of low-bit quantization can be effectively reduced. We also proposed a novel convolutional algorithm using shift operations to replace multiplication to help to deploy the GSNQ quantized models on FPGAs. Quantization comparison experiments performed on LeNet-5, AlexNet, VGG-Net, ResNet, and GoogLeNet showed that GSNQ has higher accuracy than most existing methods and achieves “lossless” quantization (i.e., the accuracy of the quantized CNN model is higher than the baseline) at low-bit quantization in most cases. FPGA comparison experiments showed that our convolutional algorithm does not occupy on-chip DSPs, and it also has a low comprehensive occupancy in terms of on-chip LUTs and FFs, which can effectively improve the computational parallelism, and this proves that GSNQ has good hardware-adaptation capability. This study provides theoretical and experimental support for the industrial application of CNNs.

Aiming to address the problem of moving mirror speed fluctuations in moving mirror control systems, an improved active disturbance rejection double closed-loop controller (IADR-DCLC) is proposed and verified by simulation to realize the high-performance control of a moving mirror control system. First, the mathematical model of a rotary-type voice coil motor (RT VCM) is established, and the relationship between the angular velocity of the RT VCM and the optical path scanning velocity is analyzed. Second, in order to suppress the model uncertainty and external disturbance of the system, an improved active disturbance rejection controller (IADRC) is proposed. Compared with a conventional ADRC, the tracking differentiator of the proposed IADRC is replaced with desired signal optimization (DSO), and the actual speed is introduced to the extended state observer (ESO). The IADRC is used in the position–speed double closed-loop control model. Finally, the simulation results show that the IADR-DCLC has not only a good tracking effect but also a good anti-interference ability and can meet the requirements of the moving mirror control system for the uniformity of optical-path scanning speed and accurate control of the position of the moving mirror.

In order to improve the control of a moving mirror control system and enhance the anti-interference ability of the system, active disturbance rejection control (ADRC) with parameter autotuning is proposed and applied to control a rotary voice coil motor (RVCM). Improved snake optimization (I-SO) was applied to tune and optimize ADRC’s key parameters. To obtain excellent parameters efficiently, in the population initialization phase of SO, the quality and diversity of initial solutions were improved through a chaotic elite opposition learning algorithm. In the local search phase, a sine and cosine (SC) search mode was introduced to enhance the local search ability of SO. The simulation results show that I-SO can effectively find the ideal parameters. I-SO has excellent search capability and stability. The experimental control system of a moving mirror was established, and the effectiveness of the parameters optimized by I-SO was verified. ADRC with parameter autotuning showed excellent control in the moving mirror control system, and the stability of the optical path scanning speed reached 99.2%.

In order to obtain highly accurate infrared spectra, the optical path scanning control system in a portable Fourier transform spectrometer (FTS) must be able to realize highly stable reciprocal scanning. To address the positional localization and speed fluctuation problems of optical path scanning control systems, an adaptive feedforward–nonlinear PI cascade composite control algorithm (AF-NLPI) is proposed. A physical model of an optical path scanning control system is established. Moreover, an adaptive feedforward compensator using a dynamic forgetting factor is proposed, and it was combined with a nonlinear PI cascade controller to form a composite controller. The control parameters were tuned using the atomic orbital search algorithm. Further, the simulation and experimental results demonstrate that the AF-NLPI can effectively improve the control accuracy and anti-interference ability of an optical path scanning control system in a portable FTS with high feasibility and practicality. By setting the scanning stroke of the system to 8 mm and scanning at 10 mm/s, the stability of the optical scanning speed reached 99.47% when controlled by the controller proposed in this paper, thus fulfilling the motion requirements for optical path scanning control systems.

A wide band range can cover more of the characteristic spectral lines of substances, and thus analyze the structure and composition of substances more accurately. In order to broaden the band range of spectral instruments, an ultra-wide-band Fourier transform infrared spectrometer is designed. The incident light of the spectrometer is constrained by a secondary imaging scheme, and switchable light sources and detectors are set to achieve an ultra-wide band coverage. A compact and highly stable double-moving mirror swing interferometer is adopted to generate optical path difference, and a controller is used to stabilize the swing of the moving mirrors. A distributed design of digital system integration and analog system integration is adopted to achieve a lightweight and low-power-consumption spectrometer. High-speed data acquisition and a transmission interface are applied to improve the real-time performance. Further, a series of experiments are performed to test the performance of the spectrometer. Finally, the experimental results show that the spectral range of the ultra-wide-band Fourier transform infrared spectrometer covers 0.770–200 μm, with an accurate wave number, a spectral resolution of 0.25 cm⁻¹, and a signal-to-noise ratio better than 50,000:1.