Quality Detection Of Green Pepper And Flowering Identification Of Eggplant Based On Deep Learning

Deep learning technology,as an emerging technology means that the machine can replace human work after a lot of "learning".In recent years,deep learning technology has been widely used in the planting and management of agricultural products,but its application areas mainly focus on the identification of pests and diseases,weed identification,fruit picking,etc.And there are few studies on the application of automatic identification of crop flowering period and auxiliary pollination.Achieving effective identification of the flowering period plays a vital role in improving fruit yield and quality,and can further promote the promotion of intelligent greenhouses.Traditional computer vision technology has low detection efficiency,low detection accuracy and poor model robustness,and is easily restricted by external environments and external conditions.Based on the above problems and status quo,this paper applies computer vision technology based on deep learning to agricultural production.The specific research content is as follows:(1)In order to preliminarily explore the application of deep learning in agricultural production,the convolutional neural network based on deep learning is first applied to the quality detection of green peppers to explore its advantages in the application of simple image contour recognition.Experiments have proved that the application of convolutional neural network can better complete the task of green pepper quality classification,and solve the classification difficulties caused by small size and large quantity in the process of green pepper quality inspection.Compared with traditional classification methods,it has obvious advantages.(2)In order to further improve the classification efficiency of the classification model,the transfer learning method is used,and the pre-trained feature extraction network is combined with the traditional machine learning classifier.Firstly,the original image is used as input,and the convolutional neural network is used for feature extraction,which solves the difficult problem of traditional model feature extraction;Secondly,the machine learning classification algorithm is used as a feature classifier to improve the efficiency of feature classification;Finally,using the migration method to learn the performance of the existing advanced training network as a feature extractor,through experiments on many categories of fruits and vegetables data set,the classification accuracy of 99.1%,the AUC value is 0.9996,and improve the efficiency of the model training,to enhance the model robustness and over fitting phenomenon plays an inhibitory effect.(3)In order to solve the problem of the identification and location of the flowering date in the automatic pollination of eggplant flower,enhance the pollination management of eggplant flower,and improve the efficiency of eggplant planting,a model for the identification of the flowering date of eggplant flower based on Mask R-CNN was proposed.In order to solve the problems of the lack of accuracy of the original network target identification,and the misdetection and omission of the large target,the model was improved.Firstly,the bilateral filtering algorithm is used to preprocess the image to achieve the purpose of edge preserving and denoising;Secondly,a method of fusion hybrid dilated convolution is proposed to enable the model to obtain a larger receptive field during the feature extraction process.Increase the segmentation accuracy of large targets;Finally,the use of transfer learning methods reduces the risk of overfitting the model and obtains a faster fitting speed.Using the improved model on the test set,the mAP(Mean Average Precision)is 0.962,and the mIOU(Mean Intersection over Union)is 0.715.Experiments show that the improved model can accurately identify the flowering period of eggplant flowers and segment it from the background;On the other hand,the segmentation accuracy of large objects has been significantly improved.