Search

Compared with other parts of the plant, leaves are easy to obtain and have obvious features. Therefore, leaf identification is also considered as one of the best methods for plant identification. For this reason, a transfer learning leaf recognition method based on VGG16 deep learning network is proposed. This method firstly carries out preprocess operations such as background whitening and standard normalization processing, then carries out data augmentation, then uses transfer learning to train VGG16 network model, and finally completes leaf recognition. The experimental results of Middle European Woody (MEW) plants data set and UCI Folio Leaf data set show that the recognition accuracy under a single background reaches 93.4% and 97.9%, respectively, which are significantly higher than traditional convolutional neural network and other recognition methods, effectively improving the recognition accuracy rate of plant leaves.

With the rapid development of artificial intelligence, the purpose of image classification is not only to identify the major categories of objects, but also to classify the images of the same category into more detailed subcategories. In order to effectively discriminate small differences between regions, a fine-grained classification algorithm was proposed based on Attention Self-supervision (ASS). Firstly, the ResNet-152 pre-trained model was applied to extract the global image features and to forecast the attention maps with Depthwise Separable convolution (DS conv). Subsequently, the Weakly Supervised Data Augmentation Network (WS-DAN) was applied to feed the augmented image back into the network and enhance the generalization ability of the network to prevent overfitting. Finally, linear fusion of attention features in ASS network was further performed to improve the accuracy of the classification. The experimental results show that the method achieves 89.6% accuracy rate on CUB-200-2011 dataset.

In the three-dimensional coordinate system, there are four omnidirectional antenna radar stations with fixed coordinates. These radar stations form a network radar and observe the target. In this paper, the beat signal with white Gaussian noise obtained by mixing the FMCW(Frequency Modulated Continuous Wave) transmission signal and the echo signal is processed by FFT to obtain the frequency value of the beat signal. Finally, the state information including the target distance obtained by the improved Jacobi iterative algorithm is compared and compared with the traditional Jacobi iterative algorithm. The simulation results were analyzed, and the superiority of the improved algorithm was verified, which provided help for accurate and timely meteorological observations.

Aiming at the conditions of that the quality of ATM video image easily affected under weak light environment, this paper presents two improved algorithms respectively for noise suppression and image enhancement in the video image preprocessing. Compared with traditional methods, the algorithms can get better precision and signal to noise ratio and keep further details while improving the contrast of image, so the image quality can be notably improved. The above algorithms only process the luminance component to ensure the efficiency.

The measurement based on FFT power spectrum of pulse transient dynamic response of structural silicone sealant damage in Hidden frame supported glass curtain wall was first proposed in the paper. The signal of four testing points was gathered and FFT Power Spectrum was used to analyze. Power spectral main peak frequency of the signal in contributing proportion of the total power (i ƒO ) was taken as a dependent variable of the structural silicone sealant damage length ( a ). Experimental data were fitted to establish a functional relations between i¥ a i¦ and i¥ i ƒO i¦byGaussian function or Lorentzian function, and the small damage value of i¥ a i¦was used to be detected. Research results show that it will be helpful for the application of the structural silicone sealant damage detection technology in the hidden frame supported glass curtain wall.

A novel mathematic model for a dual three-phase induction motor was proposed. A set of equivalent three-phase windings was built based on same magnetomotive force and power. A dual three-phase induction motor was substituted equivalently by a three-phase induction motor after a winding transformation. Then with the mathematical model of the three-phase phase induction motor in the stationary reference frame, the model of a dual three-phase induction motor was determined. A 4 pole dual three-phase induction motor was simulated and tested. Compared the simulation results with corresponding test results, the maximum error was lower than 2%. So this winding transformation and modeling method are validated and correct.

This paper deals with the problem of delay-dependent robust stability for uncertain linear systems with time varying multi-state delays. Some new delay-dependent stability criteria are derived by using free weighting matrices which are selected by means of linear matrix inequalities, and the new criteria are less conservative than existing ones. Finally, an example is given to illustrate the effectiveness of the proposed criteria.

A novel and simple pulse Dc bias test system has been established to emulate an isothermal environment for on-wafer characterization of MMIC power amplifiers. Testing results of a Ka band monolithic power amplifier with pulse dc bias show a power output capacity increase of 1.5-2dBm at the frequency of 33GHz

There is a lot of multimedia data types, related compression algorithms and related code standards that have different characteristics and adapt to different application circumstances. Getting the characteristics of multimedia dynamically is helpful for clearly understanding and properly using the media data. The objective of this research is to develop a network multimedia evaluation platform based on the client/server mode, by means of analyzing and comparing the feathers of main media (the series of MPEG, ITU-H, ITU-G, JPEG, etc) compression algorithm. The platform, MATEP, can test and display the main parameters of related multimedia data. These parameters can be used in QoS control.

Trellis codes (including convolutional codes and turbo codes) have been widely used in DS-CDMA systems for their good performance. In this paper we study the performance of convolutional codes and turbo codes in an environment where E/sub b//N/sub 0/ changes sinusoidally. We study four cases of both scenarios. For each case we obtain some conclusions and give their analysis. In this paper a very important characteristic, named additivity, is acquired.