Apple leaf diseases (APD) significantly impact tree health and crop yield by causing leaf damage, reducing photosynthesis, and making trees more susceptible to other infections. However, existing techniques struggle with detecting multiple diseases, lighting variations, and generalizing across different cultivation conditions. To overcome this challenge, a novel deep learning-based APSE-NET is proposed for APD detection. The input images are collected from Turkey Plant Dataset and preprocessed by Joint Bilateral Filter (JBF) to reduce the noise and enhance the image quality. The denoised images are fed into DenseGoogleNet (DeGoNet) that combines the dense connectivity of DenseNet with the architectural principles of GoogleNet (Inception modules) to efficiently capture multi-scale features. Gated Self-Attentive Convoluted MobileNetV3 (GSACMNet) is a lightweight deep learning model that integrates self-attention and gating mechanisms into MobileNetV3, enabling it to focus on important regions of the image. It effectively uses the extracted features for accurate and efficient classification. The proposed APSE-NET model is assessed based on its f1 score (F1), specificity (SP), precision (PR), recall (RE) and accuracy (AC). The proposed APSE-NET model achieves a high accuracy of 98.89% in ALD classification. Compared to CBAM, DWT and CNN the proposed model improves overall accuracy by 4.88%, 0.26% and 2.74% respectively