Table 8 Detailed ablation study showing component contribution and runtime Impact.

From: AI-driven drone technology and computer vision for early detection of crop disease in large agricultural areas

Configuration

Accuracy (%)

Precision (%)

Recall (%)

F1-Score (%)

Latency (s/image)

Throughput (images/s)

Full AgroVisionNet

94.2

93.5

92.8

93.1

1.2

45

w/o Transformer (replaced by Bi-LSTM)

91.6

91.2

90.8

90.9

1.0

52

w/o CNN (replaced by Global Average Pooling)

90.7

89.9

89.5

89.6

0.8

58

w/o Fusion Module (direct concatenation)

91.1

90.4

90.1

90.2

1.1

48