Tness of the MAF module proposed in this paper, we also employed the information set collected in the Science Park inside the west campus of China Agriculture University, such as the images of maize illnesses which include southern leaf blight, fusarium head blight, and these 3 kinds mentioned above. Additionally, we created the mobile detection device depending on the iOS platform, which won the second prize within the National Laptop Style Competition for Chinese College Students. As shown in Figure 20, the optimized model based on the proposed technique can immediately and efficiently detect maize diseases in practical application scenarios, proving the proposed model’s robustness.Figure 20. Screenshot of launch web page and detection pages.five. Conclusions This paper proposed an MAF module to optimize mainstream CNNs and gained outstanding benefits in detecting maize leaf illnesses with the accuracy reaching 97.41 on MAF-ResNet50. Compared with all the original network model, the accuracy enhanced by 2.33 . Because the CNN was unstable, non-convergent and overfitting when the image set was insufficient, a number of image pre-processing strategies, meanwhile, models have been applied to extend and augment the information of illness samples, for example DCGAN. Transfer studying and warm-up methods had been adopted to accelerate the training speed from the model. To confirm the effectiveness from the proposed strategy, this paper applied this model to numerous mainstream CNNs; the results indicated that the functionality of networks addingRemote Sens. 2021, 13,18 ofthe MAF module have all been enhanced. Afterward, this paper discussed the efficiency of distinct combinations of 5 base activation functions. Depending on a large quantity of experiments, the combination of Sigmoid, ReLU (or tanh), and Mish (or LeakReLU) reached the highest rate of accuracy, which was 97.41 . The result proved the effectiveness from the MAF module, and also the improvement is of considerable significance to agricultural production. The optimized module proposed within this paper can be well applied to many CNNs. Within the future, the author will make efforts to replace the combination of linear activation functions with that of nonlinear activation functions and make extra network parameters participate in model instruction.Author Contributions: Conceptualization, Y.Z.; methodology, Y.Z.; validation, Y.Z., X.Z.; writing– original draft preparation, Y.Z.; writing–review and editing, Y.Z., S.W.; visualization, Y.L., P.S.; supervision, Y.Z.; project administration, Y.Z.; funding acquisition, Q.M. All authors have read and agreed towards the published version from the manuscript. Funding: This work was supported by the 2021 Organic Science Fund Project in Shandong Province (ZR202102220347). Institutional Critique Board Statement: Not applicable. Informed Consent Statement: Not applicable. Information Availability Statement: Not applicable. Acknowledgments: We’re Arimoclomol Purity grateful for the ECC of CIEE in China Agricultural Reldesemtiv Technical Information University for their robust assistance for the duration of our thesis writing. We are also grateful for the emotional assistance supplied by Manzhou Li for the author Y.Z. Conflicts of Interest: The authors declare no conflict of interest.
remote sensingArticleContinuous Detection of Surface-Mining Footprint in Copper Mine Applying Google Earth EngineMaoxin Zhang 1 , Tingting He 1, , Guangyu Li two , Wu Xiao 1 , Haipeng Song 1 , Debin Luand Cifang WuDepartment of Land Management, Zhejiang University, Hangzhou 310058, China; [email protected] (M.Z.); [email protected] (W.X.); sh.