Activation functions are crucial in deep learning networks, given that the nonlinear ability of activation functions endows deep neural networks with real artificial intelligence. Nonlinear nonmonotonic activation functions, such as rectified linear units, Tan hyperbolic (tanh), Sigmoid, Swish, Mish, and Logish, perform well in deep learning models; however, only a few of them are widely used in mostly all applications due to their existing inconsistencies. Inspired by the MB-C-BSIF method, this study proposes Smish, a novel nonlinear activation function, expressed as f(x)=x·tanh[ln(1+sigmoid(x))], which could overcome other activation functions with good properties. Logarithmic operations are first used to reduce the range of sigmoid(x). The value is then calculated using the tanh operator. Inputs are ultimately used to multiply the previous value, thus exhibiting negative output regularization. Experiments show that Smish tends to operate more efficiently than Logish, Mish, and other activation functions on EfficientNet models with open datasets. Moreover, we evaluated the performance of Smish in various deep learning models and the parameters of its function f(x)=αx·tanh[ln(1+sigmoid(βx))], and where α = 1 and β = 1, Smish was found to exhibit the highest accuracy. The experimental results show that with Smish, the EfficientNetB3 network exhibits a Top-1 accuracy of 84.1% on the CIFAR-10 dataset; the EfficientNetB5 network has a Top-1 accuracy of 99.89% on the MNIST dataset; and the EfficientnetB7 network has a Top-1 accuracy of 91.14% on the SVHN dataset. These values are superior to those obtained using other state-of-the-art activation functions, which shows that Smish is more suitable for complex deep learning models.

With the wide increase in global forestry resources trade, the demand for wood is increasing day by day, especially rare wood. Finding a computer-based method that can identify wood species has strong practical value and very important significance for regulating the wood trade market and protecting the interests of all parties, which is one of the important problems to be solved by the wood industry. This article firstly studies the establishment of wood microscopic images dataset through a combination of traditional image amplification technology and Mix-up technology expansion strategy. Then with the traditional Faster Region-based Convolutional Neural Networks (Faster RCNN) model, the receptive field enhancement Spatial Pyramid Pooling (SPP) module and the multi-scale feature fusion of Feature Pyramid Networks (FPN) module are introduced to construct a microscopic image identification model based on the migration learning fusion model and analyzes the three factors (Mix-up, Enhanced SPP and FPN modules) affecting the wood microscopic image detection model. The experimental results show that the proposed approach can identify 10 kinds of wood microscopic images, and the accuracy rate has increased from 77.8% to 83.8%, which provides convenient conditions for further in-depth study of the microscopic characteristics of wood cells and is of great significance to the field of wood science.

Hyperspectral remote-sensing images have the characteristics of large transmission data and high propagation requirements, so they are faced with transmission and preservation problems in the process of transmission. In view of this situation, this paper proposes a spectral image reconstruction algorithm based on GISMT compressed sensing and interspectral prediction. Firstly, according to the high spectral correlation of hyperspectral remote-sensing images, the hyperspectral images are grouped according to the band, and a standard band is determined in each group. The standard band in each group is weighted by the GISMT compressed sensing method. Then, a prediction model of the general band in each group is established to realize the remote-sensing image reconstruction in the general band. Finally, the difference between the actual measured value and the predicted value is calculated. According to the prediction algorithm, the corresponding difference vector is obtained and the predicted measured value is iteratively updated by the difference vector until the hyperspectral reconstructed image of the relevant general band is finally reconstructed. It is shown by experiments that this method can effectively improve the reconstruction effect of hyperspectral images.

The tree sway frequency is an important part of the dynamic properties of trees. In order to obtain trees sway frequency in wind, a method of tracking and measuring the sway frequency of leafless deciduous trees by adaptive tracking window based on MOSSE was proposed. Firstly, an adaptive tracking window is constructed for the observed target. Secondly, the tracking method based on Minimum Output Sum Of Squared Error Filter (MOSSE) is used to track tree sway. Thirdly, Fast Fourier transform was used to analyze the horizontal sway velocity of the target area on the trees, and the sway frequency was determined. Finally, comparing the power spectral densities (PSDs) of the x axis acceleration measured by the accelerometer and PSDs of the x axis velocity measured by the video, the fundamental sway frequency measured by the accelerometer is equal to the fundamental sway frequency measured by video. The results show that the video-based method can be used successfully for measuring the sway frequency of leafless deciduous trees.

No abstract