Enhancing fruit disease classification with an advanced 3D shallow deep neural network for precise and efficient identification

• Published in: Expert systems with applications Vol. 293; p. 128559
• Main Authors: Khandagale, Hridaynath Pandurang, Patil, Sangram Tukaram, Gavali, Vikram Sampat, Manjrekar, Amrita Arvind, Halkarnikar, Pratap Pandurang
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.12.2025
• Subjects: 3D Shallow Deep Neural Network; Crisscross Harris Hawks Optimizer; Fruit disease; Newton Time-Extracting Wavelet Transform; Regularized Bias-Aware Ensemble Kalman Filtering; 3D Shallow Deep Neural Network; Crisscross Harris Hawks Optimizer; Fruit disease; Newton Time-Extracting Wavelet Transform; Regularized Bias-Aware Ensemble Kalman Filtering
• ISSN: 0957-4174
• DOI: 10.1016/j.eswa.2025.128559

Diseases in fruits cause major problems in the agricultural industry and lead to significant economic loss. These diseases reduce yields, degrade the variety and can lead to crops being withdrawn from cultivation. Existing methods rely on heavy-weight architectures, which require higher storage capacity and expensive training operations due to the large number of parameters involved. In this paper, Enhancing Fruit Disease Classification with an Advanced 3D Shallow Deep Neural Network for Precise and Efficient Identification (EFDDNN-EI) is proposed. Here, the input images are taken from two datasets: (i) Strawberry Diseases dataset, (ii) Apple Fruit Disease dataset. The input images from both datasets are pre-processed by Regularized Bias-Aware Ensemble Kalman Filtering (RBAEKF) to eliminate unwanted noise. The pre-processed images are fed to the Newton Time-Extracting Wavelet Transform (NTEWT) to extract relevant attributes, such as color, texture, shape, size. The extracted features are fed into the 3D Shallow Deep Neural Network (3DSDNN). The 3DSDNN classifies the input images from the Strawberry Diseases dataset as botrytis cinerea, slugs, aphids, sunburn, thrips, root rot, vegetable green insects, viral diseases, powdery mildew and healthy leaves. Similarly, images from the Apple Fruit Disease dataset are classified into categories, like Blotch Apple, Normal Apple, Rot Apple, Scab Apple. Finally, Crisscross Harris Hawks Optimizer (CCHHO) is employed to enhance EFDDNN-EI that can classify the fruit diseases by maximizing the accuracy and minimizing the computation time. The proposed EFDDNN-EI approach is implemented, and the performance metrics, like recall, f1-score, precision, testing accuracy, testing loss and ROC to assess the efficacy of the method. The EFDDNN-EI technique achieves 16.89%, 18.57% and 27.68% higher testing accuracy, and 18.81%, 26.53% and 30.62% higher recall compared to the existing approaches: Strawberry Disease Detection under Transfer Learning of Deep Convolutional Neural Networks (SDD-TLDCNN), A Better, Lightweight YOLOV5 Algorithm for Strawberry Disease Identification (SDD-YOLOV5), and Structural Invariant Feature Segmentation dependent Apple Fruit Disease Detection utilizing Deep Spectral Generative Adversarial Networks (SIFFD-DSGAN).