Table 9 Percentage performance degradation when removing individual AgroVisionNet Components.

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

Component removed

Accuracy drop (%)

Precision drop (%)

Recall drop (%)

F1-score drop (%)

Transformer layers (replaced by 2-layer Bi-LSTM)

2.4

2.3

2.2

2.2

CNN feature extraction (replaced by global average pooling)

4.7

4.7

4.5

4.6

IoT sensor integration (replaced by direct concatenation)

4.1

4.0

3.8

3.8

Edge optimisation (TFLite/quantisation disabled)

2.2

2.1

2.0

2.0

Back to article page