<scp>HACDRAN</scp>: A Novel Hybrid Atrous Convolution Model With Densenet and Residual Attenuation Network for Multi‐Plant Leaf Disease Classification

ABSTRACT The beneficiation of the plant is majorly crucial for both the environment and human life. The crops do trouble from disorders like other species. Various plant leaf disorders happen and trouble the general growth of the crop. These crop leaf disorders are troubles the entire plant consisting flower, root, stem, and leaf. The plant leaf disorders are mostly not taken care of by the farmers so that the crop dies or can create the cause of fruits, flowers and leave a drop. Significant detection of these disorders is essential for detecting and takes preventive measures of plant leaf disorders. The research of plant leaf disorders, procedures, and their causes for handling and managing is known as plant pathology. However, the conventional model encloses the involvement of humans in the identification and categorization of crop leaf disorders. This approach is expensive and consumes more time. It is highly essential to design a new multi‐plant leaf disorder categorization model to tackle the above‐mentioned complexities. At first, the multi‐plant leaf pictures are obtained from online sources and they are offered to the abnormality segmentation phase. Here, the abnormality presented in the multi‐plant leaf image is segmented by adaptive TransUNet, where the parameters are optimized by the developed Mean Position of Sheep Flock and Cuckoo Search (MPSFCS). Further, the abnormality segmented images are offered to the multi‐plant leaf disease classification phase. In this phase, the multi‐plant leaf disorders are categorized by utilizing Hybrid Atrous Convolution based Networks of DenseNet and Residual Attenuation Network (HACDRAN), and their parameters are optimized by the developed MPSFCS to offer effective multi‐plant leaf disease classified outcome. Thus, the developed multi‐plant leaf disease classification model provides a better functionality rate than the conventional methods with multiple analyses.