To eliminate the noise and temperature drift in an Micro-Electro-Mechanical Systems (MEMS) gyroscope’s output signal for improving measurement accuracy, a parallel processing model based on Multi-objective particle swarm optimization based on variational modal decomposition-time-frequency peak filter (MOVMD–TFPF) and Beetle antennae search algorithm- Elman neural network (BAS–Elman NN) is established. Firstly, variational mode decomposition (VMD) is optimized by multi-objective particle swarm optimization (MOPSO); then, the best decomposition parameters [kbest,abest] can be obtained. Secondly, the gyroscope output signals are decomposed by VMD optimized by MOPSO (MOVMD); then, the intrinsic mode functions (IMFs) obtained after decomposition are classified into a noise segment, mixed segment, and drift segment by sample entropy (SE). According to the idea of a parallel model, the noise segment can be discarded directly, the mixed segment is denoised by time-frequency peak filtering (TFPF), and the drift segment is compensated at the same time. In the compensation part, the beetle antennae search algorithm (BAS) is adopted to optimize the network parameters of the Elman neural network (Elman NN). Subsequently, the double-input/single-output temperature compensation model based on the BAS-Elman NN is established to compensate the drift segment, and these processed segments are reconstructed to form the final gyroscope output signal. Experimental results demonstrate the superiority of this parallel processing model; the angle random walk of the compensated gyroscope output is decreased from 0.531076 to 5.22502 × 10−3°/h/√Hz, and its bias stability is decreased from 32.7364°/h to 0.140403°/h, respectively.

Traditional fruit and vegetable classification is mostly based on manual operation, which is inefficient. Deep convolution neural network shows excellent performance in feature learning and expression. In this paper, an automatic recognition system of fruits and vegetables based on deep convolution neural network is designed. By using depthwise separable convolution instead of the traditional standard convolution, a neural network is constructed with less parameters, which is suitable for equipment with limited resources. A small data set including 12 kinds of common fruits and 8 kinds of common vegetables is established for training and testing through network download and physical shooting. The experimental results show that the recognition accuracy reaches 95.67%.

No abstract

No abstract

Traditional fruit recognition is mainly manual, which is not conducive to automation. Deep convolution neural network (DCNN) has a strong ability of feature learning and expression. It is helpful to realize intelligence in fruit sales market if it is applied to the identification of fruits. Due to the lack of standard image databases and various types of fruits, image data sets used in this paper are obtained through taking pictures of fruits and network download. Considering the small number of samples, in addition to using common image processing technology for data expansion, transfer learning technology based on vgg16 model is also adopted for fine tuning, which can reduce the training time and alleviate over fitting. Finally, six kinds of common fruits are chosen for experiments, and the test results show that the average recognition accuracy reaches 94.16%.