Cervical cancer is one of the leading causes of mortality among women, emphasizing the need for accurate diagnostic methods particularly in developing countries where access to regular screening is limited. Early detection and accurate classification of cervical cancer stages are crucial for effective treatment and improved survival rates.This study explores the potential of deep learning based convolutional neural networks (CNNs) for classifying cervical cytological images from the Sipahan Kanker Metadata (SIPaKMeD) dataset.The SIPaKMeD dataset originally containing 4,049 images is augmented to 24,294 images to enhance model generalization. We employed VGG-16, EfficientNet-B7, and CapsNet CNN models using transfer learning with ImageNet pretrained weights to improve classification accuracy.The experimental results show that EfficientNet-B7 achieved the average highest classification accuracy of 91.34%, outperforming VGG-16 (86.5%) and CapsNet (81.34%). Evaluation metrics such as precision, recall, and F1-score further validate the robustness of EfficientNet-B7 in distinguishing between different cervical cancer stages. After testing with various hyperparameters, EfficientNet-B7 minimizes misclassification errors and is able to categorize data more accurately compared to other CNN models.These findings highlight the potential of deep learning CNNs for automated cervical cancer diagnosis, aiding doctors in clinical decision-making to classify medical images and diagnose diseases. Consequently, diagnostic accuracy improves, facilitating more effective treatment planning in the healthcare sector.

Growth of the farmer is by the good yield of crops. Crop growth can be predicted by observing the physical condition of the leaves, stem etc., of the plant. This paper mainly focuses on extraction of leaf images using image segmentation technique on the digital plant images. The segmented leaf images can be used for analysing plant characteristics. In the proposed methodology, a very first step is image enhancement using statistical based morphological operations. The second step involves finding the tiny white noises and can be removed by using Erode, then the green contours are extracted by the simple method of chain approximation, and the attached objects can be separated. The required object is retained after separating the objects. Finally using transformation distance find the veins, separate veins using kernel tophat transformation and on contors bases grow regions to identify and segment each leaf. The proposed method achieves good accuracy when plant leaves are not overlapped at maximum extent.