Research on Deep Learning-based Cross-disciplinary Applications

Li, Zhuo; Meng, Lin

doi:10.1109/icamechs57222.2022.10003391

Cited by 5 publications

(4 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recently, some methods in computer vision have changed from traditional pattern recognition to deep learning. The deep learning model represented by the convolutional neural network (CNN) significantly improves the accuracy of recognition problems by automatically learning a hierarchical feature representation from raw data [6]. Therefore, significant progress has been made in crop identification and pest detection [7,8].…”

Section: Introductionmentioning

confidence: 99%

“…Pruning algorithm based on 2EFPInput: The pre-trained model weight is { l } L l =1 , the number of filters per layer{ Ñl } L l =1 and Ñl = N l × (1 − p l ), l ∈ {1, 2, … , L}, the iterate is t ∈ {1, 2, … , T }.Output: The model weight after pruning is { l } L l =1 and l ∈ ℝ Ñl × Ml ×K l ×K l 1: for all 1 ≤ l ≤ L do 2: Calculate the expected 2D entropy of feature map by Equation(5). Get the importance evaluation of each filter  l by Equation(6). Critical filters l are saved through the pruning rate p l .…”

mentioning

confidence: 99%

See 1 more Smart Citation

Real‐time seed sorting system via 2D information entropy‐based CNN pruning and TensorRt acceleration

Liao

et al. 2023

IET Image Processing

View full text Add to dashboard Cite

Seed sorting based on deep neural networks is one of the important applications of seed variety identification and quality purification. However, DNNs is difficult to deploy on embedded devices since the consumption of computational and storage resource. To address these problems, this paper proposes a pipeline‐style neural network framework for real‐time seed sorting. First, we propose a novel algorithm, 2D information entropy, pruning redundant filters to realize structured pruning. Then, the pruning rate of each convolution layer is determined by visualizing the results of 2D entropy. Meanwhile, the pruned network is fine‐tuned to recover the performance. Finally, TensorRT is utilized to optimize and accelerate the pruned model for deployment in Jeston Nano. Experiments on two large‐scale seed‐sorting datasets demonstrate the significant improvement of the proposed method over existing model compression methods. Experimental results on Jeston Nano show that the pruned model 2EFP‐E achieves a single image inference speed of 107 FPS, with the best accuracy of 95.94% on the red kidney bean dataset.

show abstract

Section: Introductionmentioning

confidence: 99%

mentioning

confidence: 99%

Real‐time seed sorting system via 2D information entropy‐based CNN pruning and TensorRt acceleration

Liao

et al. 2023

IET Image Processing

View full text Add to dashboard Cite

show abstract

“…As a study using operator theory, a nonlinear temperature control system has been designed for an aluminum plate with a Peltier device that has uncertainty [10]. In recent years, due to the popularity of AI research [11][12][13][14][15][16][17][18][19], research combining operator theory and AI has also been conducted. In addition, research on multi-input/output systems, such as research on a two-dimensional direction-of-arrival estimation using a polarization rectangular array under multipath propagation [20], has attracted much attention and has been actively studied in operator theory.…”

Section: Introductionmentioning

confidence: 99%

MSVR & Operator-Based System Design of Intelligent MIMO Sensorless Control for Microreactor Devices

Kato,

Nishizawa,

Deng

2023

Computation

View full text Add to dashboard Cite

Recently, microreactors, which are tubular reactors capable of fast and highly efficient chemical reactions, have attracted attention. However, precise temperature control is required because temperature changes due to reaction heat can cause reactions to proceed differently from those designed. In a previous study, a single-input/output nonlinear control system was proposed using a model in which the microreactor is divided into three regions and the thermal equation is formulated considering the temperature gradient, but it could not control two different temperatures. In this paper, a multi-input, multi-output nonlinear control system was designed using operator theory. On the other hand, when the number of parallel microreactors is increased, a sensorless control method using M–SVR with a generalized Gaussian kernel was incorporated into the MIMO nonlinear control system from the viewpoint of cost reduction, and the effectiveness of the proposed method was confirmed via experimental results.

show abstract

“…Convolution neural networks (CNNs) have been proven to be effective in various applications [1][2][3]: object detection [4,5], cultural heritage protection [6], environment monitoring [7], robotics [8][9][10] and healthcare [11].…”

Section: Introductionmentioning

confidence: 99%

Deep Learning Architecture Improvement Based on Dynamic Pruning and Layer Fusion

Meng

2023

Electronics

Self Cite

View full text Add to dashboard Cite

The heavy workload of current deep learning architectures significantly impedes the application of deep learning, especially on resource-constrained devices. Pruning has provided a promising solution to compressing the bloated deep learning models by removing the redundancies of the networks. However, existing pruning methods mainly focus on compressing the superfluous channels without considering layer-level redundancies, which results in the channel-pruned models still suffering from serious redundancies. To mitigate this problem, we propose an effective compression algorithm for deep learning models that uses both the channel-level and layer-level compression techniques to optimize the enormous deep learning models. In detail, the channels are dynamically pruned first, and then the model is further optimized by fusing the redundant layers. Only a minor performance loss results. The experimental results show that the computations of ResNet-110 are reduced by 80.05%, yet the accuracy is only decreased by 0.72%. Forty-eight convolutional layers could be discarded from ResNet-110 with no loss of performance, which fully demonstrates the efficiency of the proposal.

show abstract

Research on Deep Learning-based Cross-disciplinary Applications

Cited by 5 publications

References 26 publications

Real‐time seed sorting system via 2D information entropy‐based CNN pruning and TensorRt acceleration

Real‐time seed sorting system via 2D information entropy‐based CNN pruning and TensorRt acceleration

MSVR & Operator-Based System Design of Intelligent MIMO Sensorless Control for Microreactor Devices

Deep Learning Architecture Improvement Based on Dynamic Pruning and Layer Fusion

Contact Info

Product

Resources

About