Crop Disease Recognition Based on Improved ShuffleNetV2 Model
In view of the complex bottom feature extraction of traditional digital image processing technology,the recognition method of convolutional neural network has many parameters,large amount of calculation,complex network structure and low prac-ticability,this paper aims to build a crop disease recognition model that can be used in mobile applications,and takes three leaf im-ages of apple scab,apple rust and apple health as the research object.From the two dimensions of improving accuracy and reducing computation,a disease recognition model based on improved shuffleNetV2 convolutional neural network is proposed:1)Embed-ding sk attention mechanism.2)Convolution kernel with extended depth separable convolution.3)Clipping useless convolution.4)Using the prelu activation function instead.The experimental results show that the accuracy of the improved model in the data set of APLE_Mix is 98.75%,which is 2.05%higher than that of the original shuffleNetV2,the amount of flops calculation is reduced by 18.9%,the amount of parameters is increased by 6.9%,and the memory is increased by 0.03 MB(all are within acceptable range),which can better balance the model complexity and recognition accuracy.
万方数据
维普
