Research And Application Of Strawberry Common Disease Identification Based On Deep Learning

The growth of strawberries will be stressed by biological or abiotic factors,which will pose a great threat to the yield and quality of strawberries.The most important ones are various strawberry diseases.However,traditional identification methods have high misjudgment rates and poor real-time performance.In today’s era of increasing demand for strawberry yield and quality,it is clear that traditional strawberry disease identification methods that mainly rely on personal experience and visual observation cannot satisfy people’s needs Demand for disease identification and prevention.This requires finding a more effective method to efficiently identify strawberry diseases,and provide corresponding disease description and control methods.As a new and vigorous development direction in the field of machine learning,deep learning has been widely used in the field of image identification.For the image identification of common diseases of strawberry,it belongs to the fine-grained image classification,and the goal is to identify different diseases of the crop of strawberry,which has wide application value.Therefore,it is of high social value to seek a common strawberry disease identification method that can solve the above problems accurately,non-destructively,quickly and conveniently.Based on deep learning technology,this paper conducts relevant research on the identification of strawberry common diseases,and the innovations achieved are as follows:(1)On multiple agricultural technology service platforms,use the Scrapy web crawler to crawl strawberry disease images uploaded by strawberry growers,thereby constructing Strawberry Common Diseases Image Dataset(SCDID),including 15 kinds of strawberry common diseases.At present,there are no publicly available image datasets in the field of strawberry diseases.For the SCDID dataset,it can be used not only for the research in this article,but also for further publication and publicity,thereby promoting the development of strawberry disease identification.(2)This paper proposes a fine-grained fine-tuning classification model of strawberry common disease images based on the attention mechanism.A novel attention mechanism is mainly designed,which can effectively use the informative regions of the image and use transfer learning to quickly establish a fine-grained classification model of common strawberry diseases based on the attention mechanism.The results show that the attention mechanism can improve the accuracy of strawberry disease classification by 1.06%.(3)In this paper,a self-supervised multi-network collaboration strawberry common disease image fine-grained classification model is proposed.By using the self-supervised mechanism,the disease regions of the strawberry disease image can be effectively recognized without manual labeling(such as bounding box).The model consists of three networks,including the Location network,the Feedback network and the Classification network.The model enables the Location network to detect diseased regions under the guidance of the Feedback network,while the Classification network recognizes and classifies the proposed diseased regions.The model is compared with the pre-trained fine-tuning classification model for common strawberry disease images,the fine-grained fine-tuning classification model for strawberry common disease images based on the attention mechanism,and the model that achieves the best classification effect on the CUB-200-2011 dataset.The model can achieve the best identification and classification effect on the SCDID,with classification accuracy of 92.48%.In addition,the two fine-grained classification models designed in this paper are not only applicable to the SCDID used in this study,but also have generality,such as tomato,potato and other crop datasets.(4)In the research,a mini program for recognizing common strawberry diseases based on We Chat platform was developed.At present,there are few studies on strawberry disease identification based on mobile applications.Based on the self-supervised multi-network collaboration strawberry common disease image fine-grained classification model,a mini program for strawberry common disease identification based on WeChat platform is designed and implemented.After testing,the mini program identification accuracy is 90.23%.In addition,this mini program runs on We Chat platform,and users can quickly obtain accurate identification results of strawberry diseases,corresponding disease descriptions and prevention measures through simple steps.In practice,this mini program has high guiding value and will provide important support for intelligent research and application of strawberry disease identification in China.