Plant Disease Detection: PyramidNet‐ICNN Architecture With Modified BIRCH Segmentation

ABSTRACT Agriculture stands as the primary occupation in India, yet it faces a substantial annual loss of 35% in crop productivity due to plant diseases. These diseases pose a significant task in the sector of agriculture, emphasising the critical need for their automatic identification to efficiently monitor plant health. The conventional technique of analysis by specialists in laboratories is costly and time‐consuming, even though the signs of the majority of diseases appear in plant leaves. Recognising the vital importance of early issue identification, this research proposes a novel hybrid Architecture, a hybrid of PyramidNet and ICNN models (Py‐ICNN) for plant disease detection and classification with an Improved BIRCH (I‐BIRCH) segmentation model, which uses an image as input. This framework follows a systematic approach, comprising preprocessing, segmentation, extraction of features and detection and classification of diseases. Using median and Contrast Limited Adaptive Histogram Equalisation (CLAHE) filtering, the input image first undergoes enhanced preprocessing. The preprocessed outcome is then subjected to I‐Balanced Iterative Reducing and Clustering Using Hierarchies (BIRCH) segmentation. Then, features including IPHOG, multi‐texton features and MBP‐based features are extracted from the segmented image. These extracted features are then individually processed using PyramidNet and improved convolutional neural network (ICNN) to detect and classify the plant disease. Furthermore, the proposed Py‐ICNN model is evaluated and compared with traditional methods. The findings demonstrate that the Py‐ICNN framework obtained an accuracy of 93.70% and a specificity of 95.82%. These results demonstrate how well the Py‐ICNN approach detects and classifies plant diseases.