Crop Disease Detection against Complex Background Based on Improved Atrous Spatial Pyramid Pooling

Timely crop disease detection, pathogen identification, and infestation severity assessments can aid prevention control efforts to mitigate crop-yield decline. However, improved monitoring methods are needed that extract high-resolution, accurate, rich color spatial features from leaf spots in the field achieve precise fine-grained disease-severity classification sensitive disease-recognition accuracy. Here, we propose a neural-network-based method incorporating an Rouse pyramid pooling strategy detection against complex background. For neural network construction, first, dual-attention module was introduced into cross-stage partial backbone enable extraction of multi-dimensional information channel space perspectives. Next, dilated convolution-based integrated within broaden scope collection crop-disease-related images crops field. The tested using set sample data constructed collected at rate 40 frames per second occupied only 17.12 MB storage space. Field analysis conducted miniaturized model revealed average precision approaching 90.15% exceeded corresponding rates obtained comparable conventional methods. Collectively, these results indicate proposed simplified tasks suppressed noise transmission greater accuracy than is obtainable similar methods, thus demonstrating should be suitable for use practical applications related recognition.