MobileNetV3 Architecture Enhanced with Multi-Head Self-Attention for Precise Tomato Maturity Detection

To maximize agricultural output and guarantee crop quality, accurate and automated tomato maturity identification is essential. In order to improve feature extraction and classification accuracy for tomato maturity detection, we provide an Attention-Boosted MobileNetV3 architecture coupled with a Multi-Head Self-Attention (MHSA) mechanism in this work. The improved MobileNetV3 model incorporates MHSA to capture contextual linkages and long-range dependencies inside images, while utilizing its lightweight structure for computational performance. Challenges like the variation in tomato size, shape, and colour across different phases of maturation are successfully addressed by this hybrid technique. An extensive set of studies on a variety of tomato picture datasets showed that the suggested model outperformed traditional CNN designs. The findings show notable gains in recall, accuracy, precision, and $\mathbf{F} 1$-score, which makes it a good fit for real-time agricultural applications. The suggested approach opens the door for developments in automated harvesting and smart agriculture by providing a scalable and reliable solution for intelligent crop monitoring systems.